aid stringlengths 9 15 | mid stringlengths 7 10 | abstract stringlengths 78 2.56k | related_work stringlengths 92 1.77k | ref_abstract dict |
|---|---|---|---|---|
1902.05546 | 2913988491 | Contemporary sensorimotor learning approaches typically start with an existing complex agent (e.g., a robotic arm), which they learn to control. In contrast, this paper investigates a modular co-evolution strategy: a collection of primitive agents learns to dynamically self-assemble into composite bodies while also learning to coordinate their behavior to control these bodies. Each primitive agent consists of a limb with a motor attached at one end. Limbs may choose to link up to form collectives. When a limb initiates a link-up action and there is another limb nearby, the latter is magnetically connected to the 'parent' limb's motor. This forms a new single agent, which may further link with other agents. In this way, complex morphologies can emerge, controlled by a policy whose architecture is in explicit correspondence with the morphology. We evaluate the performance of these 'dynamic' and 'modular' agents in simulated environments. We demonstrate better generalization to test-time changes both in the environment, as well as in the agent morphology, compared to static and monolithic baselines. Project videos and code are available at this https URL | Encoding graphical structures into neural networks has been used for a large number of applications, including visual question answering @cite_28 , quantum chemistry @cite_5 , semi-supervised classification @cite_12 , and representation learning @cite_15 . The works most similar to ours involve learning control policies @cite_21 @cite_16 . For example, Nervenet @cite_21 represents individual limbs and joints as nodes in a graph and demonstrates multi-limb generalization, just like our system does. However, the morphologies on which Nervenet operates are not learned jointly with the policy and hand-defined to be compositional in nature. Others @cite_25 @cite_26 have shown that graph neural networks can also be applied to inference models as well as to planning. Many of these past works implement some variant of Graph Neural Networks @cite_10 which operate on general graphs. Our method leverages the constraint that the morphologies can always be represented as a rooted tree in order to simplify the message passing. | {
"cite_N": [
"@cite_26",
"@cite_28",
"@cite_21",
"@cite_5",
"@cite_15",
"@cite_16",
"@cite_10",
"@cite_25",
"@cite_12"
],
"mid": [
"2815036811",
"2416885651",
"2785948534",
"2952254971",
"2808344987",
"2804072623",
"2116341502",
"2805516822",
"2519887557"
],
"abstract": [
"Our goal is to generate a policy to complete an unseen task given just a single video demonstration of the task in a given domain. We hypothesize that to successfully generalize to unseen complex tasks from a single video demonstration, it is necessary to explicitly incorporate the compositional structure of the tasks into the model. To this end, we propose Neural Task Graph (NTG) Networks, which use conjugate task graph as the intermediate representation to modularize both the video demonstration and the derived policy. We empirically show NTG achieves inter-task generalization on two complex tasks: Block Stacking in BulletPhysics and Object Collection in AI2-THOR. NTG improves data efficiency with visual input as well as achieve strong generalization without the need for dense hierarchical supervision. We further show that similar performance trends hold when applied to real-world data. We show that NTG can effectively predict task structure on the JIGSAWS surgical dataset and generalize to unseen tasks.",
"Visual question answering is fundamentally compositional in nature---a question like \"where is the dog?\" shares substructure with questions like \"what color is the dog?\" and \"where is the cat?\" This paper seeks to simultaneously exploit the representational capacity of deep networks and the compositional linguistic structure of questions. We describe a procedure for constructing and learning *neural module networks*, which compose collections of jointly-trained neural \"modules\" into deep networks for question answering. Our approach decomposes questions into their linguistic substructures, and uses these structures to dynamically instantiate modular networks (with reusable components for recognizing dogs, classifying colors, etc.). The resulting compound networks are jointly trained. We evaluate our approach on two challenging datasets for visual question answering, achieving state-of-the-art results on both the VQA natural image dataset and a new dataset of complex questions about abstract shapes.",
"We address the problem of learning structured policies for continuous control. In traditional reinforcement learning, policies of agents are learned by MLPs which take the concatenation of all observations from the environment as input for predicting actions. In this work, we propose NerveNet to explicitly model the structure of an agent, which naturally takes the form of a graph. Specifically, serving as the agent's policy network, NerveNet first propagates information over the structure of the agent and then predict actions for different parts of the agent. In the experiments, we first show that our NerveNet is comparable to state-of-the-art methods on standard MuJoCo environments. We further propose our customized reinforcement learning environments for benchmarking two types of structure transfer learning tasks, i.e., size and disability transfer. We demonstrate that policies learned by NerveNet are significantly better than policies learned by other models and are able to transfer even in a zero-shot setting.",
"Supervised learning on molecules has incredible potential to be useful in chemistry, drug discovery, and materials science. Luckily, several promising and closely related neural network models invariant to molecular symmetries have already been described in the literature. These models learn a message passing algorithm and aggregation procedure to compute a function of their entire input graph. At this point, the next step is to find a particularly effective variant of this general approach and apply it to chemical prediction benchmarks until we either solve them or reach the limits of the approach. In this paper, we reformulate existing models into a single common framework we call Message Passing Neural Networks (MPNNs) and explore additional novel variations within this framework. Using MPNNs we demonstrate state of the art results on an important molecular property prediction benchmark; these results are strong enough that we believe future work should focus on datasets with larger molecules or more accurate ground truth labels.",
"Modern deep transfer learning approaches have mainly focused on learning generic feature vectors from one task that are transferable to other tasks, such as word embeddings in language and pretrained convolutional features in vision. However, these approaches usually transfer unary features and largely ignore more structured graphical representations. This work explores the possibility of learning generic latent relational graphs that capture dependencies between pairs of data units (e.g., words or pixels) from large-scale unlabeled data and transferring the graphs to downstream tasks. Our proposed transfer learning framework improves performance on various tasks including question answering, natural language inference, sentiment analysis, and image classification. We also show that the learned graphs are generic enough to be transferred to different embeddings on which the graphs have not been trained (including GloVe embeddings, ELMo embeddings, and task-specific RNN hidden unit), or embedding-free units such as image pixels.",
"Understanding and interacting with everyday physical scenes requires rich knowledge about the structure of the world, represented either implicitly in a value or policy function, or explicitly in a transition model. Here we introduce a new class of learnable models--based on graph networks--which implement an inductive bias for object- and relation-centric representations of complex, dynamical systems. Our results show that as a forward model, our approach supports accurate predictions from real and simulated data, and surprisingly strong and efficient generalization, across eight distinct physical systems which we varied parametrically and structurally. We also found that our inference model can perform system identification. Our models are also differentiable, and support online planning via gradient-based trajectory optimization, as well as offline policy optimization. Our framework offers new opportunities for harnessing and exploiting rich knowledge about the world, and takes a key step toward building machines with more human-like representations of the world.",
"Many underlying relationships among data in several areas of science and engineering, e.g., computer vision, molecular chemistry, molecular biology, pattern recognition, and data mining, can be represented in terms of graphs. In this paper, we propose a new neural network model, called graph neural network (GNN) model, that extends existing neural network methods for processing the data represented in graph domains. This GNN model, which can directly process most of the practically useful types of graphs, e.g., acyclic, cyclic, directed, and undirected, implements a function tau(G,n) isin IRm that maps a graph G and one of its nodes n into an m-dimensional Euclidean space. A supervised learning algorithm is derived to estimate the parameters of the proposed GNN model. The computational cost of the proposed algorithm is also considered. Some experimental results are shown to validate the proposed learning algorithm, and to demonstrate its generalization capabilities.",
"Artificial intelligence (AI) has undergone a renaissance recently, making major progress in key domains such as vision, language, control, and decision-making. This has been due, in part, to cheap data and cheap compute resources, which have fit the natural strengths of deep learning. However, many defining characteristics of human intelligence, which developed under much different pressures, remain out of reach for current approaches. In particular, generalizing beyond one's experiences--a hallmark of human intelligence from infancy--remains a formidable challenge for modern AI. The following is part position paper, part review, and part unification. We argue that combinatorial generalization must be a top priority for AI to achieve human-like abilities, and that structured representations and computations are key to realizing this objective. Just as biology uses nature and nurture cooperatively, we reject the false choice between \"hand-engineering\" and \"end-to-end\" learning, and instead advocate for an approach which benefits from their complementary strengths. We explore how using relational inductive biases within deep learning architectures can facilitate learning about entities, relations, and rules for composing them. We present a new building block for the AI toolkit with a strong relational inductive bias--the graph network--which generalizes and extends various approaches for neural networks that operate on graphs, and provides a straightforward interface for manipulating structured knowledge and producing structured behaviors. We discuss how graph networks can support relational reasoning and combinatorial generalization, laying the foundation for more sophisticated, interpretable, and flexible patterns of reasoning. As a companion to this paper, we have released an open-source software library for building graph networks, with demonstrations of how to use them in practice.",
"We present a scalable approach for semi-supervised learning on graph-structured data that is based on an efficient variant of convolutional neural networks which operate directly on graphs. We motivate the choice of our convolutional architecture via a localized first-order approximation of spectral graph convolutions. Our model scales linearly in the number of graph edges and learns hidden layer representations that encode both local graph structure and features of nodes. In a number of experiments on citation networks and on a knowledge graph dataset we demonstrate that our approach outperforms related methods by a significant margin."
]
} |
1902.05326 | 2909795240 | Distinguishing between classes of time series sampled from dynamic systems is a common challenge in systems and control engineering, for example in the context of health monitoring, fault detection, and quality control. The challenge is increased when no underlying model of a system is known, measurement noise is present, and long signals need to be interpreted. In this paper we address these issues with a new non parametric classifier based on topological signatures. Our model learns classes as weighted kernel density estimates (KDEs) over persistent homology diagrams and predicts new trajectory labels using Sinkhorn divergences on the space of diagram KDEs to quantify proximity. We show that this approach accurately discriminates between states of chaotic systems that are close in parameter space, and its performance is robust to noise. | Time series have previously been analysed using TDA, largely via PH of point clouds constructed using delay embeddings @cite_5 @cite_42 @cite_15 @cite_37 . However delay embeddings require heuristic estimation of the embedding dimension and delay size to be computed beforehand, increase the influence of noise, and lead to a rapid increase in the complexity of computing PH of filtrations on the point cloud, requiring subsampling techniques such as the witness complex @cite_33 @cite_14 @cite_6 . One previous study avoids embeddings but relies on a coarse grained statistic derived from PDs to characterise time series, the persistent entropy @cite_3 . Our method is similar to this since we use a filtration on the time series directly, however we use a metric on the space of persistence diagrams directly rather than on the space of persistent entropy histograms. | {
"cite_N": [
"@cite_37",
"@cite_14",
"@cite_33",
"@cite_42",
"@cite_6",
"@cite_3",
"@cite_5",
"@cite_15"
],
"mid": [
"2498184556",
"1486210143",
"2963213081",
"2102368730",
"2963056332",
"",
"1798863697",
"2033896831"
],
"abstract": [
"This work introduces a new dataset and framework for the exploration of topological data analysis (TDA) techniques applied to time-series data. We examine the end-toend TDA processing pipeline for persistent homology applied to time-delay embeddings of time series – embeddings that capture the underlying system dynamics from which time series data is acquired. In particular, we consider stability with respect to time series length, the approximation accuracy of sparse filtration methods, and the discriminating ability of persistence diagrams as a feature for learning. We explore these properties across a wide range of time-series datasets spanning multiple domains for single source multi-segment signals as well as multi-source single segment signals. Our analysis and dataset captures the entire TDA processing pipeline and includes time-delay embeddings, persistence diagrams, topological distance measures, as well as kernels for similarity learning and classification tasks for a broad set of time-series data sources. We outline the TDA framework and rationale behind the dataset and provide insights into the role of TDA for time-series analysis as well as opportunities for new work.",
"This paper tackles the problem of computing topological invariants of geometric objects in a robust manner, using only point cloud data sampled from the object. It is now widely recognised that this kind of topological analysis can give qualitative information about data sets which is not readily available by other means. In particular, it can be an aid to visualisation of high dimensional data. Standard simplicial complexes for approximating the topological type of the underlying space (such as Cech, Rips, or a-shape) produce simplicial complexes whose vertex set has the same size as the underlying set of point cloud data. Such constructions are sometimes still tractable, but are wasteful (of computing resources) since the homotopy types of the underlying objects are generally realisable on much smaller vertex sets. We obtain smaller complexes by choosing a set of 'landmark' points from our data set, and then constructing a \"witness complex\" on this set using ideas motivated by the usual Delaunay complex in Euclidean space. The key idea is that the remaining (non-landmark) data points are used as witnesses to the existence of edges or simplices spanned by combinations of landmark points. Our construction generalises the topology-preserving graphs of Martinetz and Schulten [MS94] in two directions. First, it produces a simplicial complex rather than a graph. Secondly it actually produces a nested family of simplicial complexes, which represent the data at different feature scales, suitable for calculating persistent homology [ELZ00, ZC04]. We find that in addition to the complexes being smaller, they also provide (in a precise sense) a better picture of the homology, with less noise, than the full scale constructions using all the data points. We illustrate the use of these complexes in qualitatively analyzing a data set of 3 × 3 pixel patches studied by David [LPM03].",
"Topology-based analysis of time-series data from dynamical systems is powerful: it potentially allows for computer-based proofs of the existence of various classes of regular and chaotic invariant sets for high-dimensional dynamics. Standard methods are based on a cubical discretization of the dynamics and use the time series to construct an outer approximation of the underlying dynamical system. The resulting multivalued map can be used to compute the Conley index of isolated invariant sets of cubes. In this paper we introduce a discretization that uses instead a simplicial complex constructed from a witness-landmark relationship. The goal is to obtain a natural discretization that is more tightly connected with the invariant density of the time series itself. The time-ordering of the data also directly leads to a map on this simplicial complex that we call the witness map. We obtain conditions under which this witness map gives an outer approximation of the dynamics and thus can be used to compute the C...",
"We develop in this paper a theoretical framework for the topological study of time series data. Broadly speaking, we describe geometrical and topological properties of sliding window embeddings, as seen through the lens of persistent homology. In particular, we show that maximum persistence at the point-cloud level can be used to quantify periodicity at the signal level, prove structural and convergence theorems for the resulting persistence diagrams, and derive estimates for their dependency on window size and embedding dimension. We apply this methodology to quantifying periodicity in synthetic data sets and compare the results with those obtained using state-of-the-art methods in gene expression analysis. We call this new method SW1PerS, which stands for Sliding Windows and 1-Dimensional Persistence Scoring.",
"Topological data analysis (TDA), while abstract, allows a characterization of time-series data obtained from nonlinear and complex dynamical systems. Though it is surprising that such an abstract measure of structure—counting pieces and holes—could be useful for real-world data, TDA lets us compare different systems, and even do membership testing or change-point detection. However, TDA is computationally expensive and involves a number of free parameters. This complexity can be obviated by coarse-graining, using a construct called the witness complex. The parametric dependence gives rise to the concept of persistent homology: how shape changes with scale. Its results allow us to distinguish time-series data from different systems—e.g., the same note played on different musical instruments.",
"",
"Abstract This paper describes a new approach for ascertaining the stability of stochastic dynamical systems in their parameter space by examining their time series using topological data analysis (TDA). We illustrate the approach using a nonlinear delayed model that describes the tool oscillations due to self-excited vibrations in turning. Each time series is generated using the Euler-Maruyama method and a corresponding point cloud is obtained using the Takens embedding. The point cloud can then be analyzed using a tool from TDA known as persistent homology. The results of this study show that the described approach can be used for analyzing datasets of delay dynamical systems generated both from numerical simulation and experimental data. The contributions of this paper include presenting for the first time a topological approach for investigating the stability of a class of nonlinear stochastic delay equations, and introducing a new application of TDA to machining processes.",
"We explore computational topology for clustering.We focus on time series and spatial data.Shape features are extracted based on n-dimensional holes.We run experiments with synthetic and real-world data.Results show improvements when compared to traditional clustering. Topology is the branch of mathematics that studies how objects relate to one another for their qualitative structural properties, such as connectivity and shape. In this paper, we present an approach for data clustering based on topological features computed over the persistence diagram, estimated using the theory of persistent homology. The features indicate topological properties such as Betti numbers, i.e., the number of n-dimensional holes in the discretized data space. The main contribution of our approach is enabling the clustering of time series that have similar recurrent behavior characterized by their attractors in phase space and spatial data that have similar scale-invariant spatial distributions, as traditional clustering techniques ignore that information as they rely on point-to-point dissimilarity measures such as Euclidean distance or elastic measures. We present experiments that confirm the usefulness of our approach with time series and spatial data applications in the fields of biology, medicine and ecology."
]
} |
1902.05326 | 2909795240 | Distinguishing between classes of time series sampled from dynamic systems is a common challenge in systems and control engineering, for example in the context of health monitoring, fault detection, and quality control. The challenge is increased when no underlying model of a system is known, measurement noise is present, and long signals need to be interpreted. In this paper we address these issues with a new non parametric classifier based on topological signatures. Our model learns classes as weighted kernel density estimates (KDEs) over persistent homology diagrams and predicts new trajectory labels using Sinkhorn divergences on the space of diagram KDEs to quantify proximity. We show that this approach accurately discriminates between states of chaotic systems that are close in parameter space, and its performance is robust to noise. | Our method uses persistence images (PIs) as its underlying stable representation of PH, however rather than using these as feature vectors in a support vector classifier as in @cite_24 , we apply a distance metric to them. Following @cite_22 we treat PIs as kernel estimators of the density of expected PDs for a class, and to measure distance between these density estimates and new PIs we use the Sinkhorn divergence @cite_29 . The latter is an upper bound approximation to the Wasserstein distance between distributions which can be computed very quickly @cite_2 @cite_25 @cite_7 . The Sinkhorn divergence has recently been applied to scalable clustering and averaging of large PDs @cite_34 , but not yet to classification using topological features. | {
"cite_N": [
"@cite_22",
"@cite_7",
"@cite_29",
"@cite_24",
"@cite_2",
"@cite_34",
"@cite_25"
],
"mid": [
"2788076228",
"2009172320",
"2158131535",
"2964237352",
"2964308809",
"2804231530",
"2765207424"
],
"abstract": [
"Persistence diagrams play a fundamental role in Topological Data Analysis where they are used as topological descriptors of filtrations built on top of data. They consist in discrete multisets of points in the plane @math that can equivalently be seen as discrete measures in @math . When the data come as a random point cloud, these discrete measures become random measures whose expectation is studied in this paper. First, we show that for a wide class of filtrations, including the C ech and Rips-Vietoris filtrations, the expected persistence diagram, that is a deterministic measure on @math , has a density with respect to the Lebesgue measure. Second, building on the previous result we show that the persistence surface recently introduced in [Adams & al., Persistence images: a stable vector representation of persistent homology] can be seen as a kernel estimator of this density. We propose a cross-validation scheme for selecting an optimal bandwidth, which is proven to be a consistent procedure to estimate the density.",
"This paper introduces a new class of algorithms for optimization problems involving optimal transportation over geometric domains. Our main contribution is to show that optimal transportation can be made tractable over large domains used in graphics, such as images and triangle meshes, improving performance by orders of magnitude compared to previous work. To this end, we approximate optimal transportation distances using entropic regularization. The resulting objective contains a geodesic distance-based kernel that can be approximated with the heat kernel. This approach leads to simple iterative numerical schemes with linear convergence, in which each iteration only requires Gaussian convolution or the solution of a sparse, pre-factored linear system. We demonstrate the versatility and efficiency of our method on tasks including reflectance interpolation, color transfer, and geometry processing.",
"Optimal transport distances are a fundamental family of distances for probability measures and histograms of features. Despite their appealing theoretical properties, excellent performance in retrieval tasks and intuitive formulation, their computation involves the resolution of a linear program whose cost can quickly become prohibitive whenever the size of the support of these measures or the histograms' dimension exceeds a few hundred. We propose in this work a new family of optimal transport distances that look at transport problems from a maximum-entropy perspective. We smooth the classic optimal transport problem with an entropic regularization term, and show that the resulting optimum is also a distance which can be computed through Sinkhorn's matrix scaling algorithm at a speed that is several orders of magnitude faster than that of transport solvers. We also show that this regularized distance improves upon classic optimal transport distances on the MNIST classification problem.",
"Many data sets can be viewed as a noisy sampling of an underlying space, and tools from topological data analysis can characterize this structure for the purpose of knowledge discovery. One such tool is persistent homology, which provides a multiscale description of the homological features within a data set. A useful representation of this homological information is a persistence diagram (PD). Efforts have been made to map PDs into spaces with additional structure valuable to machine learning tasks. We convert a PD to a finite-dimensional vector representation which we call a persistence image (PI), and prove the stability of this transformation with respect to small perturbations in the inputs. The discriminatory power of PIs is compared against existing methods, showing significant performance gains. We explore the use of PIs with vector-based machine learning tools, such as linear sparse support vector machines, which identify features containing discriminating topological information. Finally, high accuracy inference of parameter values from the dynamic output of a discrete dynamical system (the linked twist map) and a partial differential equation (the anisotropic Kuramoto-Sivashinsky equation) provide a novel application of the discriminatory power of PIs.",
"Computing optimal transport distances such as the earth mover's distance is a fundamental problem in machine learning, statistics, and computer vision. Despite the recent introduction of several algorithms with good empirical performance, it is unknown whether general optimal transport distances can be approximated in near-linear time. This paper demonstrates that this ambitious goal is in fact achieved by Cuturi's Sinkhorn Distances. This result relies on a new analysis of Sinkhorn iterations, which also directly suggests a new greedy coordinate descent algorithm Greenkhorn with the same theoretical guarantees. Numerical simulations illustrate that Greenkhorn significantly outperforms the classical Sinkhorn algorithm in practice.",
"Persistence diagrams (PDs) are now routinely used to summarize the underlying topology of sophisticated data encountered in challenging learning problems. Despite several appealing properties, integrating PDs in learning pipelines can be challenging because their natural geometry is not Hilbertian. In particular, algorithms to average a family of PDs have only been considered recently and are known to be computationally prohibitive. We propose in this article a tractable framework to carry out fundamental tasks on PDs, namely evaluating distances, computing barycenters and carrying out clustering. This framework builds upon a formulation of PD metrics as optimal transport (OT) problems, for which recent computational advances, in particular entropic regularization and its convolutional formulation on regular grids, can all be leveraged to provide efficient and (GPU) scalable computations. We demonstrate the efficiency of our approach by carrying out clustering on PDs at scales never seen before in the literature.",
"Entropic regularization is quickly emerging as a new standard in optimal transport (OT). It enables to cast the OT computation as a differentiable and unconstrained convex optimization problem, which can be efficiently solved using the Sinkhorn algorithm. However, entropy keeps the transportation plan strictly positive and therefore completely dense, unlike unregularized OT. This lack of sparsity can be problematic in applications where the transportation plan itself is of interest. In this paper, we explore regularizing the primal and dual OT formulations with a strongly convex term, which corresponds to relaxing the dual and primal constraints with smooth approximations. We show how to incorporate squared @math -norm and group lasso regularizations within that framework, leading to sparse and group-sparse transportation plans. On the theoretical side, we bound the approximation error introduced by regularizing the primal and dual formulations. Our results suggest that, for the regularized primal, the approximation error can often be smaller with squared @math -norm than with entropic regularization. We showcase our proposed framework on the task of color transfer."
]
} |
1902.05326 | 2909795240 | Distinguishing between classes of time series sampled from dynamic systems is a common challenge in systems and control engineering, for example in the context of health monitoring, fault detection, and quality control. The challenge is increased when no underlying model of a system is known, measurement noise is present, and long signals need to be interpreted. In this paper we address these issues with a new non parametric classifier based on topological signatures. Our model learns classes as weighted kernel density estimates (KDEs) over persistent homology diagrams and predicts new trajectory labels using Sinkhorn divergences on the space of diagram KDEs to quantify proximity. We show that this approach accurately discriminates between states of chaotic systems that are close in parameter space, and its performance is robust to noise. | Much previous work has been done on characterising and comparing trajectories of dynamical systems, but this often focuses on methods for distinguishing between chaotic and non-chaotic (periodic, quasi-periodic, intermittent) behaviour, for example by analysing spectra of Lyapunov exponents or textures of recurrence plots @cite_35 @cite_21 . The method we develop here is suited to this type of classification but to showcase its fine-grained capabilities we focus numerical experiments on showing that it can distinguish between different chaotic regimes that lie very close to one another in parameter space. | {
"cite_N": [
"@cite_35",
"@cite_21"
],
"mid": [
"2081681829",
"2152254020"
],
"abstract": [
"Abstract Recurrence is a fundamental property of dynamical systems, which can be exploited to characterise the system's behaviour in phase space. A powerful tool for their visualisation and analysis called recurrence plot was introduced in the late 1980's. This report is a comprehensive overview covering recurrence based methods and their applications with an emphasis on recent developments. After a brief outline of the theory of recurrences, the basic idea of the recurrence plot with its variations is presented. This includes the quantification of recurrence plots, like the recurrence quantification analysis, which is highly effective to detect, e. g., transitions in the dynamics of systems from time series. A main point is how to link recurrences to dynamical invariants and unstable periodic orbits. This and further evidence suggest that recurrences contain all relevant information about a system's behaviour. As the respective phase spaces of two systems change due to coupling, recurrence plots allow studying and quantifying their interaction. This fact also provides us with a sensitive tool for the study of synchronisation of complex systems. In the last part of the report several applications of recurrence plots in economy, physiology, neuroscience, earth sciences, astrophysics and engineering are shown. The aim of this work is to provide the readers with the know how for the application of recurrence plot based methods in their own field of research. We therefore detail the analysis of data and indicate possible difficulties and pitfalls.",
"We present the first algorithms that allow the estimation of non-negative Lyapunov exponents from an experimental time series. Lyapunov exponents, which provide a qualitative and quantitative characterization of dynamical behavior, are related to the exponentially fast divergence or convergence of nearby orbits in phase space. A system with one or more positive Lyapunov exponents is defined to be chaotic. Our method is rooted conceptually in a previously developed technique that could only be applied to analytically defined model systems: we monitor the long-term growth rate of small volume elements in an attractor. The method is tested on model systems with known Lyapunov spectra, and applied to data for the Belousov-Zhabotinskii reaction and Couette-Taylor flow."
]
} |
1902.05414 | 2912717377 | Manual counts of mitotic figures, which are determined in the tumor region with the highest mitotic activity, are a key parameter of most tumor grading schemes. It is however strongly dependent on the area selection. To reduce potential variability of prognosis due to this, we propose to use an algorithmic field of interest prediction to assess the area of highest mitotic activity in a whole-slide image. Methods: We evaluated two state-of-the-art methods, all based on the use of deep convolutional neural networks on their ability to predict the mitotic count in digital histopathology slides. We evaluated them on a novel dataset of 32 completely annotated whole slide images from canine cutaneous mast cell tumors (CMCT) and one publicly available human mamma carcinoma (HMC) dataset. We first compared the mitotic counts (MC) predicted by the two models with the ground truth MC on both data sets. Second, for the CMCT data set, we compared the computationally predicted position and MC of the area of highest mitotic activity with size-equivalent areas selected by eight veterinary pathologists. Results: We found a high correlation between the mitotic count as predicted by the models (Pearson's correlation coefficient between 0.931 and 0.962 for the CMCT data set and between 0.801 and 0.986 for the HMC data set) on the slides. For the CMCT data set, this is also reflected in the predicted position representing mitotic counts in mostly the upper quartile of the slide's ground truth MC distribution. Further, we found strong differences between experts in position selection. Conclusion: While the mitotic counts in areas selected by the experts substantially varied, both algorithmic approaches were consistently able to generate a good estimate of the area of highest mitotic count. To achieve better inter-rater agreement, we propose to use computer-based area selection for manual mitotic count. | Mitotic figure detection is a known task in computer vision for more than three decades, starting with early approaches by Kaman using low resolution grayscale images. It took until the advent of deep learning technologies, first used by Cire s an @cite_26 , for acceptable results to be achieved. The models used in this approach are commonly pixel classifiers trained with images where a mitotic figure is either at the center (positive sample) or not (negative sample). In recent years, significant advances were made in this field, fostered also by several competitions held on this topic @cite_19 @cite_28 @cite_8 . Especially the introduction of deeper residual networks @cite_15 and object detection methods like Faster-RCNN @cite_21 had a large influence on current methods, e.g. the DeepMitosis framework by Li @cite_16 . | {
"cite_N": [
"@cite_26",
"@cite_8",
"@cite_28",
"@cite_21",
"@cite_19",
"@cite_15",
"@cite_16"
],
"mid": [
"22040386",
"2884988214",
"2132031490",
"",
"2122394460",
"2194775991",
"2789755511"
],
"abstract": [
"We use deep max-pooling convolutional neural networks to detect mitosis in breast histology images. The networks are trained to classify each pixel in the images, using as context a patch centered on the pixel. Simple postprocessing is then applied to the network output. Our approach won the ICPR 2012 mitosis detection competition, outperforming other contestants by a significant margin.",
"Abstract Tumor proliferation is an important biomarker indicative of the prognosis of breast cancer patients. Assessment of tumor proliferation in a clinical setting is a highly subjective and labor-intensive task. Previous efforts to automate tumor proliferation assessment by image analysis only focused on mitosis detection in predefined tumor regions. However, in a real-world scenario, automatic mitosis detection should be performed in whole-slide images (WSIs) and an automatic method should be able to produce a tumor proliferation score given a WSI as input. To address this, we organized the TUmor Proliferation Assessment Challenge 2016 (TUPAC16) on prediction of tumor proliferation scores from WSIs. The challenge dataset consisted of 500 training and 321 testing breast cancer histopathology WSIs. In order to ensure fair and independent evaluation, only the ground truth for the training dataset was provided to the challenge participants. The first task of the challenge was to predict mitotic scores, i.e., to reproduce the manual method of assessing tumor proliferation by a pathologist. The second task was to predict the gene expression based PAM50 proliferation scores from the WSI. The best performing automatic method for the first task achieved a quadratic-weighted Cohen's kappa score of κ = 0.567, 95 CI [0.464, 0.671] between the predicted scores and the ground truth. For the second task, the predictions of the top method had a Spearman's correlation coefficient of r = 0.617, 95 CI [0.581 0.651] with the ground truth. This was the first comparison study that investigated tumor proliferation assessment from WSIs. The achieved results are promising given the difficulty of the tasks and weakly-labeled nature of the ground truth. However, further research is needed to improve the practical utility of image analysis methods for this task.",
"The proliferative activity of breast tumors, which is routinely estimated by counting of mitotic figures in hematoxylin and eosin stained histology sections, is considered to be one of the most important prognostic markers. However, mitosis counting is laborious, subjective and may suffer from low inter-observer agreement. With the wider acceptance of whole slide images in pathology labs, automatic image analysis has been proposed as a potential solution for these issues. In this paper, the results from the Assessment of Mitosis Detection Algorithms 2013 (AMIDA13) challenge are described. The challenge was based on a data set consisting of 12 training and 11 testing subjects, with more than one thousand annotated mitotic figures by multiple observers. Short descriptions and results from the evaluation of eleven methods are presented. The top performing method has an error rate that is comparable to the inter-observer agreement among pathologists.",
"",
"Introduction: In the framework of the Cognitive Microscope (MICO) project, we have set up a contest about mitosis detection in images of H and E stained slides of breast cancer for the conference ICPR 2012. Mitotic count is an important parameter for the prognosis of breast cancer. However, mitosis detection in digital histopathology is a challenging problem that needs a deeper study. Indeed, mitosis detection is difficult because mitosis are small objects with a large variety of shapes, and they can thus be easily confused with some other objects or artefacts present in the image. We added a further dimension to the contest by using two different slide scanners having different resolutions and producing red-green-blue (RGB) images, and a multi-spectral microscope producing images in 10 different spectral bands and 17 layers Z-stack. 17 teams participated in the study and the best team achieved a recall rate of 0.7 and precision of 0.89. Context: Several studies on automatic tools to process digitized slides have been reported focusing mainly on nuclei or tubule detection. Mitosis detection is a challenging problem that has not yet been addressed well in the literature. Aims: Mitotic count is an important parameter in breast cancer grading as it gives an evaluation of the aggressiveness of the tumor. However, consistency, reproducibility and agreement on mitotic count for the same slide can vary largely among pathologists. An automatic tool for this task may help for reaching a better consistency, and at the same time reducing the burden of this demanding task for the pathologists. Subjects and Methods: Professor Frederique Capron team of the pathology department at Pitie-Salpetriere Hospital in Paris, France, has selected a set of five slides of breast cancer. The slides are stained with H and E. They have been scanned by three different equipments: Aperio ScanScope XT slide scanner, Hamamatsu NanoZoomer 2.0‑HT slide scanner and 10 bands multispectral microscope. The data set is made up of 50 high power fields (HPF) coming from 5 different slides scanned at ×40 magnification. There are 10 HPFs slide. The pathologist has annotated all the mitotic cells manually. A HPF has a size of 512 µm × 512 µm (that is an area of 0.262 mm 2 , which is a surface equivalent to that of a microscope field diameter of 0.58 mm. These 50 HPFs contain a total of 326 mitotic cells on images of both scanners, and",
"Deeper neural networks are more difficult to train. We present a residual learning framework to ease the training of networks that are substantially deeper than those used previously. We explicitly reformulate the layers as learning residual functions with reference to the layer inputs, instead of learning unreferenced functions. We provide comprehensive empirical evidence showing that these residual networks are easier to optimize, and can gain accuracy from considerably increased depth. On the ImageNet dataset we evaluate residual nets with a depth of up to 152 layers—8× deeper than VGG nets [40] but still having lower complexity. An ensemble of these residual nets achieves 3.57 error on the ImageNet test set. This result won the 1st place on the ILSVRC 2015 classification task. We also present analysis on CIFAR-10 with 100 and 1000 layers. The depth of representations is of central importance for many visual recognition tasks. Solely due to our extremely deep representations, we obtain a 28 relative improvement on the COCO object detection dataset. Deep residual nets are foundations of our submissions to ILSVRC & COCO 2015 competitions1, where we also won the 1st places on the tasks of ImageNet detection, ImageNet localization, COCO detection, and COCO segmentation.",
"Abstract Mitotic count is a critical predictor of tumor aggressiveness in the breast cancer diagnosis. Nowadays mitosis counting is mainly performed by pathologists manually, which is extremely arduous and time-consuming. In this paper, we propose an accurate method for detecting the mitotic cells from histopathological slides using a novel multi-stage deep learning framework. Our method consists of a deep segmentation network for generating mitosis region when only a weak label is given (i.e., only the centroid pixel of mitosis is annotated), an elaborately designed deep detection network for localizing mitosis by using contextual region information, and a deep verification network for improving detection accuracy by removing false positives. We validate the proposed deep learning method on two widely used Mitosis Detection in Breast Cancer Histological Images (MITOSIS) datasets. Experimental results show that we can achieve the highest F-score on the MITOSIS dataset from ICPR 2012 grand challenge merely using the deep detection network. For the ICPR 2014 MITOSIS dataset that only provides the centroid location of mitosis, we employ the segmentation model to estimate the bounding box annotation for training the deep detection network. We also apply the verification model to eliminate some false positives produced from the detection model. By fusing scores of the detection and verification models, we achieve the state-of-the-art results. Moreover, our method is very fast with GPU computing, which makes it feasible for clinical practice."
]
} |
1902.05414 | 2912717377 | Manual counts of mitotic figures, which are determined in the tumor region with the highest mitotic activity, are a key parameter of most tumor grading schemes. It is however strongly dependent on the area selection. To reduce potential variability of prognosis due to this, we propose to use an algorithmic field of interest prediction to assess the area of highest mitotic activity in a whole-slide image. Methods: We evaluated two state-of-the-art methods, all based on the use of deep convolutional neural networks on their ability to predict the mitotic count in digital histopathology slides. We evaluated them on a novel dataset of 32 completely annotated whole slide images from canine cutaneous mast cell tumors (CMCT) and one publicly available human mamma carcinoma (HMC) dataset. We first compared the mitotic counts (MC) predicted by the two models with the ground truth MC on both data sets. Second, for the CMCT data set, we compared the computationally predicted position and MC of the area of highest mitotic activity with size-equivalent areas selected by eight veterinary pathologists. Results: We found a high correlation between the mitotic count as predicted by the models (Pearson's correlation coefficient between 0.931 and 0.962 for the CMCT data set and between 0.801 and 0.986 for the HMC data set) on the slides. For the CMCT data set, this is also reflected in the predicted position representing mitotic counts in mostly the upper quartile of the slide's ground truth MC distribution. Further, we found strong differences between experts in position selection. Conclusion: While the mitotic counts in areas selected by the experts substantially varied, both algorithmic approaches were consistently able to generate a good estimate of the area of highest mitotic count. To achieve better inter-rater agreement, we propose to use computer-based area selection for manual mitotic count. | Even though results as achieved by Li on the 2012 ICPR MITOS data set @cite_19 with F1-scores of up to @math are impressive, they will likely still not meet clinical requirements, and also might have acceptance problems with experts, since robustness to factors like staining differences and image quality have not yet been proven. Additionally, the data sets overestimate the prevalence of mitotic figures, as large parts of the tumor were excluded from annotation and partially high power fields without any mitotic figure were excluded @cite_8 . | {
"cite_N": [
"@cite_19",
"@cite_8"
],
"mid": [
"2122394460",
"2884988214"
],
"abstract": [
"Introduction: In the framework of the Cognitive Microscope (MICO) project, we have set up a contest about mitosis detection in images of H and E stained slides of breast cancer for the conference ICPR 2012. Mitotic count is an important parameter for the prognosis of breast cancer. However, mitosis detection in digital histopathology is a challenging problem that needs a deeper study. Indeed, mitosis detection is difficult because mitosis are small objects with a large variety of shapes, and they can thus be easily confused with some other objects or artefacts present in the image. We added a further dimension to the contest by using two different slide scanners having different resolutions and producing red-green-blue (RGB) images, and a multi-spectral microscope producing images in 10 different spectral bands and 17 layers Z-stack. 17 teams participated in the study and the best team achieved a recall rate of 0.7 and precision of 0.89. Context: Several studies on automatic tools to process digitized slides have been reported focusing mainly on nuclei or tubule detection. Mitosis detection is a challenging problem that has not yet been addressed well in the literature. Aims: Mitotic count is an important parameter in breast cancer grading as it gives an evaluation of the aggressiveness of the tumor. However, consistency, reproducibility and agreement on mitotic count for the same slide can vary largely among pathologists. An automatic tool for this task may help for reaching a better consistency, and at the same time reducing the burden of this demanding task for the pathologists. Subjects and Methods: Professor Frederique Capron team of the pathology department at Pitie-Salpetriere Hospital in Paris, France, has selected a set of five slides of breast cancer. The slides are stained with H and E. They have been scanned by three different equipments: Aperio ScanScope XT slide scanner, Hamamatsu NanoZoomer 2.0‑HT slide scanner and 10 bands multispectral microscope. The data set is made up of 50 high power fields (HPF) coming from 5 different slides scanned at ×40 magnification. There are 10 HPFs slide. The pathologist has annotated all the mitotic cells manually. A HPF has a size of 512 µm × 512 µm (that is an area of 0.262 mm 2 , which is a surface equivalent to that of a microscope field diameter of 0.58 mm. These 50 HPFs contain a total of 326 mitotic cells on images of both scanners, and",
"Abstract Tumor proliferation is an important biomarker indicative of the prognosis of breast cancer patients. Assessment of tumor proliferation in a clinical setting is a highly subjective and labor-intensive task. Previous efforts to automate tumor proliferation assessment by image analysis only focused on mitosis detection in predefined tumor regions. However, in a real-world scenario, automatic mitosis detection should be performed in whole-slide images (WSIs) and an automatic method should be able to produce a tumor proliferation score given a WSI as input. To address this, we organized the TUmor Proliferation Assessment Challenge 2016 (TUPAC16) on prediction of tumor proliferation scores from WSIs. The challenge dataset consisted of 500 training and 321 testing breast cancer histopathology WSIs. In order to ensure fair and independent evaluation, only the ground truth for the training dataset was provided to the challenge participants. The first task of the challenge was to predict mitotic scores, i.e., to reproduce the manual method of assessing tumor proliferation by a pathologist. The second task was to predict the gene expression based PAM50 proliferation scores from the WSI. The best performing automatic method for the first task achieved a quadratic-weighted Cohen's kappa score of κ = 0.567, 95 CI [0.464, 0.671] between the predicted scores and the ground truth. For the second task, the predictions of the top method had a Spearman's correlation coefficient of r = 0.617, 95 CI [0.581 0.651] with the ground truth. This was the first comparison study that investigated tumor proliferation assessment from WSIs. The achieved results are promising given the difficulty of the tasks and weakly-labeled nature of the ground truth. However, further research is needed to improve the practical utility of image analysis methods for this task."
]
} |
1902.05414 | 2912717377 | Manual counts of mitotic figures, which are determined in the tumor region with the highest mitotic activity, are a key parameter of most tumor grading schemes. It is however strongly dependent on the area selection. To reduce potential variability of prognosis due to this, we propose to use an algorithmic field of interest prediction to assess the area of highest mitotic activity in a whole-slide image. Methods: We evaluated two state-of-the-art methods, all based on the use of deep convolutional neural networks on their ability to predict the mitotic count in digital histopathology slides. We evaluated them on a novel dataset of 32 completely annotated whole slide images from canine cutaneous mast cell tumors (CMCT) and one publicly available human mamma carcinoma (HMC) dataset. We first compared the mitotic counts (MC) predicted by the two models with the ground truth MC on both data sets. Second, for the CMCT data set, we compared the computationally predicted position and MC of the area of highest mitotic activity with size-equivalent areas selected by eight veterinary pathologists. Results: We found a high correlation between the mitotic count as predicted by the models (Pearson's correlation coefficient between 0.931 and 0.962 for the CMCT data set and between 0.801 and 0.986 for the HMC data set) on the slides. For the CMCT data set, this is also reflected in the predicted position representing mitotic counts in mostly the upper quartile of the slide's ground truth MC distribution. Further, we found strong differences between experts in position selection. Conclusion: While the mitotic counts in areas selected by the experts substantially varied, both algorithmic approaches were consistently able to generate a good estimate of the area of highest mitotic count. To achieve better inter-rater agreement, we propose to use computer-based area selection for manual mitotic count. | Lozanski have shown that inter-rater concordance of grades for count of centroblasts significantly benefits from a preselection of HPF @cite_3 . Fauzi have shown that computer-aided preselection systems for HPF can increase overall reader accuracy @cite_13 in follicular lymphoma grading based on density of centroblasts. We thus propose to predict the area of highest mitotic activity in a complete WSI to aid the human expert in finding a more accurate prognosis. | {
"cite_N": [
"@cite_13",
"@cite_3"
],
"mid": [
"2198024083",
"2099218265"
],
"abstract": [
"Background Follicular lymphoma (FL) is one of the most common lymphoid malignancies in the western world. FL cases are stratified into three histological grades based on the average centroblast count per high power field (HPF). The centroblast count is performed manually by the pathologist using an optical microscope and hematoxylin and eosin (H&E) stained tissue section. Although this is the current clinical practice, it suffers from high inter- and intra-observer variability and is vulnerable to sampling bias.",
"Context: Pathologists grade follicular lymphoma (FL) cases by selecting 10, random high power fields (HPFs), counting the number of centroblasts (CBs) in these HPFs under the microscope and then calculating the average CB count for the whole slide. Previous studies have demonstrated that there is high inter-reader variability among pathologists using this methodology in grading. Aims: The objective of this study was to explore if newly available digital reading technologies can reduce inter-reader variability. Settings and Design: In this study, we considered three different reading conditions (RCs) in grading FL: (1) Conventional (glass-slide based) to establish the baseline, (2) digital whole slide viewing, (3) digital whole slide viewing with selected HPFs. Six board-certified pathologists from five different institutions read 17 FL slides in these three different RCs. Results: Although there was relative poor consensus in conventional reading, with lack of consensus in 41.2 of cases, which was similar to previously reported studies; we found that digital reading with pre-selected fields improved the inter-reader agreement, with only 5.9 lacking consensus among pathologists. Conclusions: Digital whole slide RC resulted in the worst concordance among pathologists while digital whole slide reading selected HPFs improved the concordance. Further studies are underway to determine if this performance can be sustained with a larger dataset and our automated HPF and CB detection algorithms can be employed to further improve the concordance."
]
} |
1902.05320 | 2913541399 | Hash functions like SHA-1 or MD5 are one of the most important cryptographic primitives, especially in the field of information integrity. Considering the fact that increasing methods have been proposed to break these hash algorithms, a competition for a new family of hash functions was held by the US National Institute of Standards and Technology. Keccak was the winner and selected to be the next generation of hash function standard, named SHA-3. We aim to implement and optimize Batch mode based Keccak algorithms on NVIDIA GPU platform. Our work consider the case of processing multiple hash tasks at once and implement the case on CPU and GPU respectively. Our experimental results show that GPU performance is significantly higher than CPU is the case of processing large batches of small hash tasks. | Lowden @cite_3 focus on the exploration and analysis of the Keccak tree hashing mode on a GPU platform. Based on the implementation, there are core features of the GPU that could be used to accelerate the time it takes to complete a hash due to the massively parallel architecture of the device. In addition to analyzing the speed of the algorithm, the underlying hardware is profiled to identify the bottlenecks that limited the hash speed. The results of their work show that tree hashing can hash data at rates of up to 3 GB s for the fixed size tree mode. | {
"cite_N": [
"@cite_3"
],
"mid": [
"1862268400"
],
"abstract": [
"In an effort to provide security and data integrity, hashing algorithms have been designed to consume an input of any length to produce a fixed length output. KECCAK was selected by NIST to become the next Secure Hashing Algorithm (SHA-3) after nearly five years of competition. In addition to providing a sequential operating mode, there is also a tree mode that allows large input messages to be hashed in parallel. This work focuses on the exploration and analysis of the KECCAK tree hashing mode on a GPU platform. Based on the implementation, there are core features of the GPU that could be used to accelerate the time it takes to complete a hash due to the massively parallel architecture of the device. In addition to analyzing the speed of the algorithm, the underlying hardware is profiled to identify the bottlenecks that limited the hash speed. The results of this work show that tree hashing can hash data at rates of up to 3 GB s for the fixed size tree mode. On a 3.40 GHz CPU, this is the equivalent of 1.03 cycles per byte, more than six times faster than a sequential implementation for a very large input. For the variable size tree mode, the throughput was 500 MB s. Based on the performance analysis, modification of the input rate of the KECCAK sponge resulted in a negligible change to the overall speed. As a result of the hardware profiling, the register and L1 cache usage in the GPU was a major bottleneck to the overall throughput."
]
} |
1902.05320 | 2913541399 | Hash functions like SHA-1 or MD5 are one of the most important cryptographic primitives, especially in the field of information integrity. Considering the fact that increasing methods have been proposed to break these hash algorithms, a competition for a new family of hash functions was held by the US National Institute of Standards and Technology. Keccak was the winner and selected to be the next generation of hash function standard, named SHA-3. We aim to implement and optimize Batch mode based Keccak algorithms on NVIDIA GPU platform. Our work consider the case of processing multiple hash tasks at once and implement the case on CPU and GPU respectively. Our experimental results show that GPU performance is significantly higher than CPU is the case of processing large batches of small hash tasks. | , @cite_11 propose a GPU based AES implementation. In their implementation, the frequently accessed T-boxes were allocated on on-chip shared memory and the granularity that one thread handles a 16 Bytes AES block was adopted. Finally, they achieve a performance of around 60 Gbps throughput on NVIDIA Tesla C2050 GPU, which runs up to 50 times faster than a sequential implementation based on Intel Core i7-920 2.66GHz CPU. | {
"cite_N": [
"@cite_11"
],
"mid": [
"1965357124"
],
"abstract": [
"GPU is continuing its trend of vastly outperforming CPU while becoming more general purpose. In order to improve the efficiency of AES algorithm, this paper proposed a CUDA implementation of Electronic Codebook (ECB) mode encoding process and Cipher Feedback (CBC) mode decoding process on GPU. In our implementation, the frequently accessed T-boxes were allocated on on-chip shared memory and the granularity that one thread handles a 16 Bytes AES block was adopted. Finally, we achieved the highest performance of around 60 Gbps throughput on NVIDIA Tesla C2050 GPU, which runs up to 50 times faster than a sequential implementation based on Intel Core i7-920 2.66GHz CPU. In addition, we discussed the optimization under some practical application scenarios such as overlapping GPU processing and data transfer."
]
} |
1902.05320 | 2913541399 | Hash functions like SHA-1 or MD5 are one of the most important cryptographic primitives, especially in the field of information integrity. Considering the fact that increasing methods have been proposed to break these hash algorithms, a competition for a new family of hash functions was held by the US National Institute of Standards and Technology. Keccak was the winner and selected to be the next generation of hash function standard, named SHA-3. We aim to implement and optimize Batch mode based Keccak algorithms on NVIDIA GPU platform. Our work consider the case of processing multiple hash tasks at once and implement the case on CPU and GPU respectively. Our experimental results show that GPU performance is significantly higher than CPU is the case of processing large batches of small hash tasks. | , @cite_4 develop G-BLASTN, a GPU-accelerated nucleotide alignment tool based on the widely used NCBI-BLAST. G-BLASTN can produce exactly the same results as NCBI-BLAST, and it has very similar user commands. Compared with the sequential NCBI-BLAST, G-BLASTN can achieve an overall speedup of 14.80X under ‘megablast’ mode. They @cite_5 also propose to exploit the computing power of Graphic Processing Units (GPUs) for homomorphic hashing. Specifically, they demonstrate how to use NVIDIA GPUs and the Computer Unified Device Architecture (CUDA) programming model to achieve 38 times of speedup over the CPU counterpart. They also develop a multi-precision modular arithmetic library on CUDA platform, which is not only key to our specific application, but also very useful for a large number of cryptographic applications. | {
"cite_N": [
"@cite_5",
"@cite_4"
],
"mid": [
"1975244381",
"2163932369"
],
"abstract": [
"Multiple-precision integer operations are key components of many security applications; but unfortunately they are computationally expensive on contemporary CPUs. In this paper, we present our design and implementation of a multiple-precision integer library for GPUs which is implemented by CUDA. We report our experimental results which show that a significant speedup can be achieved by GPUs as compared with the GNU MP library on CPUs.",
"Motivation: Since 1990, the basic local alignment search tool (BLAST) has become one of the most popular and fundamental bioinformatics tools for sequence similarity searching, receiving extensive attention from the research community. The two pioneering papers on BLAST have received over 96 000 citations. Given the huge population of BLAST users and the increasing size of sequence databases, an urgent topic of study is how to improve the speed. Recently, graphics processing units (GPUs) have been widely used as low-cost, highperformance computing platforms. The existing GPU-BLAST is a promising software tool that uses a GPU to accelerate protein sequence alignment. Unfortunately, there is still no GPU-accelerated software tool for BLAST-based nucleotide sequence alignment. Results: We developed G-BLASTN, a GPU-accelerated nucleotide alignment tool based on the widely used NCBI-BLAST. G-BLASTN can produce exactly the same results as NCBI-BLAST, and it has very similar user commands. Compared with the sequential NCBIBLAST, G-BLASTN can achieve an overall speedup of 14.80X under ‘megablast’ mode. More impressively, it achieves an overall speedup of 7.15X over the multithreaded NCBI-BLAST running on 4 CPU cores. When running under ‘blastn’ mode, the overall speedups are 4.32X (against 1-core) and 1.56X (against 4-core). G-BLASTN also supports a pipeline mode that further improves the overall performance by up to 44 when handling a batch of queries as a whole. Currently G-BLASTN is best optimized for databases with long sequences. We plan to optimize its performance on short database sequences"
]
} |
1902.05320 | 2913541399 | Hash functions like SHA-1 or MD5 are one of the most important cryptographic primitives, especially in the field of information integrity. Considering the fact that increasing methods have been proposed to break these hash algorithms, a competition for a new family of hash functions was held by the US National Institute of Standards and Technology. Keccak was the winner and selected to be the next generation of hash function standard, named SHA-3. We aim to implement and optimize Batch mode based Keccak algorithms on NVIDIA GPU platform. Our work consider the case of processing multiple hash tasks at once and implement the case on CPU and GPU respectively. Our experimental results show that GPU performance is significantly higher than CPU is the case of processing large batches of small hash tasks. | , @cite_2 implements a high speed hash function Keccak (SHA3-512) using the integrated development environment CUDA for GPU is proposed. In addition, the safety level of Keccak is also discussed at the point of Pre-Image Resistance especially. In order to implement a high speed hash function for password cracking, the special program is also developed for passwords up to 71 characters. Moreover, the throughput of 2-time hash is also evaluated in their work. | {
"cite_N": [
"@cite_2"
],
"mid": [
"2782604938"
],
"abstract": [
"In August 2015, SHA-3 Standard was published as the next generation of Secure Hashing Algorithm. Several optimized implementations of SHA-3 on a 24-core processor with software and hardware co-design are proposed in this paper. For software designs, a 4-core mapping scheme with shared-memory reduces the delay time by exploring the internal parallelism of the algorithm; the scheme of 5 cores with packet router improves flexibility; and 24-core mapping with circuit router fully explores the parallelism with higher energy efficiency. Hardware approaches: specific instructions and novel latch-based accelerators, are proposed to improve performance and energy efficiency. The three software mappings provide a speedup of 3.1, 3.6 and 19.6, respectively. Almost 2 times better performance and energy efficiency are obtained with three specific instructions. The implementation of 4 latch-based accelerators achieves a throughput of 81.9 Gbps at 850MHz with 3.7 pJ bit energy consumption per bit."
]
} |
1902.05320 | 2913541399 | Hash functions like SHA-1 or MD5 are one of the most important cryptographic primitives, especially in the field of information integrity. Considering the fact that increasing methods have been proposed to break these hash algorithms, a competition for a new family of hash functions was held by the US National Institute of Standards and Technology. Keccak was the winner and selected to be the next generation of hash function standard, named SHA-3. We aim to implement and optimize Batch mode based Keccak algorithms on NVIDIA GPU platform. Our work consider the case of processing multiple hash tasks at once and implement the case on CPU and GPU respectively. Our experimental results show that GPU performance is significantly higher than CPU is the case of processing large batches of small hash tasks. | , @cite_12 contribute to the cryptography research community by presenting techniques to accelerate symmetric block ciphers (IDEA, Blowfish and Threefish) in NVIDIA GTX 690 with Kepler architecture. The results are benchmarked against implementation in OpenMP and existing GPU implementations in the literature. We are able to achieve encryption throughput of 90.3 Gbps, 50.82 Gbps and 83.71 Gbps for IDEA, Blowfish and Threefish respectively. Block ciphers can be used as pseudorandom number generator (PRNG) when it is operating under counter mode (CTR), but the speed is usually slower compare to other PRNG using lighter operations. Hence, they attempt to modify IDEA and Blowfish in order to achieve faster PRNG generation. The modified IDEA and Blowfish manage to pass all NIST Statistical Test and TestU01 Small Crush except the more stringent tests in TestU01 (Crush and BigCrush). | {
"cite_N": [
"@cite_12"
],
"mid": [
"1659252663"
],
"abstract": [
"GPU is widely used in various applications that require huge computational power. In this paper, we contribute to the cryptography research community by presenting techniques to accelerate symmetric block ciphers (IDEA, Blowfish and Threefish) in NVIDIA GTX 690 with Kepler architecture. The results are benchmarked against implementation in OpenMP and existing GPU implementations in the literature. We are able to achieve encryption throughput of 90.3 Gbps, 50.82 Gbps and 83.71 Gbps for IDEA, Blowfish and Threefish respectively. Block ciphers can be used as pseudorandom number generator (PRNG) when it is operating under counter mode (CTR), but the speed is usually slower compare to other PRNG using lighter operations. Hence, we attempt to modify IDEA and Blowfish in order to achieve faster PRNG generation. The modified IDEA and Blowfish manage to pass all NIST Statistical Test and TestU01 Small Crush except the more stringent tests in TestU01 (Crush and BigCrush)."
]
} |
1902.05400 | 2914788946 | An active area of research is to increase the safety of self-driving vehicles. Although safety cannot be guarenteed completely, the capability of a vehicle to predict the future trajectories of its surrounding vehicles could help ensure this notion of safety to a greater deal. We cast the trajectory forecast problem in a multi-time step forecasting problem and develop a Convolutional Neural Network based approach to learn from trajectory sequences generated from completely raw dataset in real-time. Results show improvement over baselines. | Several recent studies have addressed the problem of vehicle trajectory forecasting. A survey of previous work could be found in @cite_10 . Previous approaches can be grouped into two major classes. The first class of approaches treat the trajectory forecasting as a time series classification task where the aim is to assign class labels such as turn left, right or stay in lane for the trajectories. For such classification, Markov based models @cite_23 and Support Vector Machines have been adopted @cite_1 . Moreover, Neural Networks were also used in this direction where a probabilistic multilayer perceptron based approach is proposed in @cite_3 to model how likely a vehicle is to follow a particular trajectory or a lane for a given input of vehicle position history. Furthermore, the Long short term memory (LSTM) network was also investigated in @cite_18 where a framework for activity classification of on-road vehicles using 3D trajectory cues was proposed. @cite_9 also follows a classification objective where an LSTM model is trained with sequence data to produce softmax probabilities of future vehicle position on the grid. | {
"cite_N": [
"@cite_18",
"@cite_9",
"@cite_1",
"@cite_3",
"@cite_23",
"@cite_10"
],
"mid": [
"2567297805",
"2608536805",
"2012849708",
"2507412229",
"2153323685",
"2097545165"
],
"abstract": [
"Behavior analysis of vehicles surrounding the egovehicle is an essential component in safe and pleasant autonomous driving. This study develops a framework for activity classification of observed on-road vehicles using 3D trajectory cues and a Long Short Term Memory (LSTM) model. As a case study, we aim to classify maneuvers of surrounding vehicles at four way intersections. LIDAR, GPS, and IMU measurements are used to extract ego-motion compensated surround trajectories from data clips in the KITTI benchmark. The impact of different prediction label space choices, feature space input, noisy missing trajectory data, and LSTM model architectures are analyzed, presenting the strengths and limitations of the proposed approach.",
"In this paper, we propose an efficient vehicle trajectory prediction framework based on recurrent neural network. Basically, the characteristic of the vehicle's trajectory is different from that of regular moving objects since it is affected by various latent factors including road structure, traffic rules, and driver's intention. Previous state of the art approaches use sophisticated vehicle behavior model describing these factors and derive the complex trajectory prediction algorithm, which requires a system designer to conduct intensive model optimization for practical use. Our approach is data-driven and simple to use in that it learns complex behavior of the vehicles from the massive amount of trajectory data through deep neural network model. The proposed trajectory prediction method employs the recurrent neural network called long short-term memory (LSTM) to analyze the temporal behavior and predict the future coordinate of the surrounding vehicles. The proposed scheme feeds the sequence of vehicles' coordinates obtained from sensor measurements to the LSTM and produces the probabilistic information on the future location of the vehicles over occupancy grid map. The experiments conducted using the data collected from highway driving show that the proposed method can produce reasonably good estimate of future trajectory.",
"Predicting driver behavior is a key component for Advanced Driver Assistance Systems (ADAS). In this paper, a novel approach based on Support Vector Machine and Bayesian filtering is proposed for online lane change intention prediction. The approach uses the multiclass probabilistic outputs of the Support Vector Machine as an input to the Bayesian filter, and the output of the Bayesian filter is used for the final prediction of lane changes. A lane tracker integrated in a passenger vehicle is used for real-world data collection for the purpose of training and testing. Data from different drivers on different highways were used to evaluate the robustness of the approach. The results demonstrate that the proposed approach is able to predict driver intention to change lanes on average 1.3 seconds in advance, with a maximum prediction horizon of 3.29 seconds.",
"For safe and reliable autonomous driving systems, prediction of surrounding vehicles' future behavior and potential risks are critical. The state-of-the-art prediction algorithms tend to show limited performance on long-term predictions due to their deterministic nature. In this paper, a probabilistic lateral motion prediction algorithm is proposed based on multilayer perceptron (MLP) approach. The MLP model consists of two parts; target lane and trajectory models. In order to develop an intuitive and accurate prediction algorithm, a lane-based trajectory prediction model is introduced based on the fact that vehicles drive within a lane except for during lane changes. More specifically, a set of three representative trajectories with different levels of lane-change positions are generated for each target lane, and real-world traffic data is categorized by each trajectory for MLP training. These target lane and trajectory models enable the stochastic MLP modeling and training. The proposed MLP model outputs probabilities of how likely a vehicle will follow each trajectory and each lane for a given input of vehicle position history including current position. For training the MLP model, Next Generation Simulation traffic data are used. Simulation results show that the proposed algorithm detects lane-changes one to one and a half second earlier than existing methods and three seconds before lane crossing with about ninety percentages accuracy.",
"The complexity of situations occurring at intersections is demanding on the cognitive abilities of drivers. Advanced Driver Assistance Systems (ADAS) are intended to assist particularly in those situations. However, for adequate system reaction strategies it is essential to develop situation assessment. Especially the driver's intention has to be estimated. So, the criticality can be inferred and efficient intervention strategies can take action. In this paper, we present a prediction framework based on Hidden Markov Models (HMMs) and analyze its performance using a large database of real driving data. Our focus is on the variation of the model parameters and the choice of the dataset for learning. The direction of travel while approaching a 4-way intersection is to be estimated. A solid prediction is accomplished with high prediction rates above 90 and mean prediction times up to 7 seconds before entering the intersection area.",
"With the objective to improve road safety, the automotive industry is moving toward more “intelligent” vehicles. One of the major challenges is to detect dangerous situations and react accordingly in order to avoid or mitigate accidents. This requires predicting the likely evolution of the current traffic situation, and assessing how dangerous that future situation might be. This paper is a survey of existing methods for motion prediction and risk assessment for intelligent vehicles. The proposed classification is based on the semantics used to define motion and risk. We point out the tradeoff between model completeness and real-time constraints, and the fact that the choice of a risk assessment method is influenced by the selected motion model."
]
} |
1902.05551 | 2914362802 | We propose a new policy iteration theory as an important extension of soft policy iteration and Soft Actor-Critic (SAC), one of the most efficient model free algorithms for deep reinforcement learning. Supported by the new theory, arbitrary entropy measures that generalize Shannon entropy, such as Tsallis entropy and Renyi entropy, can be utilized to properly randomize action selection while fulfilling the goal of maximizing expected long-term rewards. Our theory gives birth to two new algorithms, i.e., Tsallis entropy Actor-Critic (TAC) and Renyi entropy Actor-Critic (RAC). Theoretical analysis shows that these algorithms can be more effective than SAC. Moreover, they pave the way for us to develop a new Ensemble Actor-Critic (EAC) algorithm in this paper that features the use of a bootstrap mechanism for deep environment exploration as well as a new value-function based mechanism for high-level action selection. Empirically we show that TAC, RAC and EAC can achieve state-of-the-art performance on a range of benchmark control tasks, outperforming SAC and several cutting-edge learning algorithms in terms of both sample efficiency and effectiveness. | This paper studies the collective use of three RL frameworks, i.e. the @cite_14 , the @cite_6 @cite_4 @cite_34 @cite_19 and the @cite_38 @cite_37 . Existing works such as SAC have considered the first two frameworks. It is the first time in the literature for us to further incorporate the third framework for effective environment exploration and reliable RL. | {
"cite_N": [
"@cite_38",
"@cite_37",
"@cite_14",
"@cite_4",
"@cite_6",
"@cite_19",
"@cite_34"
],
"mid": [
"2963938771",
"2043806097",
"2012587148",
"2554120691",
"2593044849",
"2962902376",
"2949561945"
],
"abstract": [
"Efficient exploration remains a major challenge for reinforcement learning (RL). Common dithering strategies for exploration, such as '-greedy, do not carry out temporally-extended (or deep) exploration; this can lead to exponentially larger data requirements. However, most algorithms for statistically efficient RL are not computationally tractable in complex environments. Randomized value functions offer a promising approach to efficient exploration with generalization, but existing algorithms are not compatible with nonlinearly parameterized value functions. As a first step towards addressing such contexts we develop bootstrapped DQN. We demonstrate that bootstrapped DQN can combine deep exploration with deep neural networks for exponentially faster learning than any dithering strategy. In the Arcade Learning Environment bootstrapped DQN substantially improves learning speed and cumulative performance across most games.",
"This paper describes several ensemble methods that combine multiple different reinforcement learning (RL) algorithms in a single agent. The aim is to enhance learning speed and final performance by combining the chosen actions or action probabilities of different RL algorithms. We designed and implemented four different ensemble methods combining the following five different RL algorithms: Q-learning, Sarsa, actor-critic (AC), QV-learning, and AC learning automaton. The intuitively designed ensemble methods, namely, majority voting (MV), rank voting, Boltzmann multiplication (BM), and Boltzmann addition, combine the policies derived from the value functions of the different RL algorithms, in contrast to previous work where ensemble methods have been used in RL for representing and learning a single value function. We show experiments on five maze problems of varying complexity; the first problem is simple, but the other four maze tasks are of a dynamic or partially observable nature. The results indicate that the BM and MV ensembles significantly outperform the single RL algorithms.",
"Policy search is a subfield in reinforcement learning which focuses on finding good parameters for a given policy parametrization. It is well suited for robotics as it can cope with high-dimensional state and action spaces, one of the main challenges in robot learning. We review recent successes of both model-free and model-based policy search in robot learning.Model-free policy search is a general approach to learn policies based on sampled trajectories. We classify model-free methods based on their policy evaluation strategy, policy update strategy, and exploration strategy and present a unified view on existing algorithms. Learning a policy is often easier than learning an accurate forward model, and, hence, model-free methods are more frequently used in practice. However, for each sampled trajectory, it is necessary to interact with the robot, which can be time consuming and challenging in practice. Model-based policy search addresses this problem by first learning a simulator of the robot's dynamics from data. Subsequently, the simulator generates trajectories that are used for policy learning. For both model-free and model-based policy search methods, we review their respective properties and their applicability to robotic systems.",
"Policy gradient is an efficient technique for improving a policy in a reinforcement learning setting. However, vanilla online variants are on-policy only and not able to take advantage of off-policy data. In this paper we describe a new technique that combines policy gradient with off-policy Q-learning, drawing experience from a replay buffer. This is motivated by making a connection between the fixed points of the regularized policy gradient algorithm and the Q-values. This connection allows us to estimate the Q-values from the action preferences of the policy, to which we apply Q-learning updates. We refer to the new technique as 'PGQL', for policy gradient and Q-learning. We also establish an equivalency between action-value fitting techniques and actor-critic algorithms, showing that regularized policy gradient techniques can be interpreted as advantage function learning algorithms. We conclude with some numerical examples that demonstrate improved data efficiency and stability of PGQL. In particular, we tested PGQL on the full suite of Atari games and achieved performance exceeding that of both asynchronous advantage actor-critic (A3C) and Q-learning.",
"We establish a new connection between value and policy based reinforcement learning (RL) based on a relationship between softmax temporal value consistency and policy optimality under entropy regularization. Specifically, we show that softmax consistent action values correspond to optimal entropy regularized policy probabilities along any action sequence, regardless of provenance. From this observation, we develop a new RL algorithm, Path Consistency Learning (PCL), that minimizes a notion of soft consistency error along multi-step action sequences extracted from both on- and off-policy traces. We examine the behavior of PCL in different scenarios and show that PCL can be interpreted as generalizing both actor-critic and Q-learning algorithms. We subsequently deepen the relationship by showing how a single model can be used to represent both a policy and the corresponding softmax state values, eliminating the need for a separate critic. The experimental evaluation demonstrates that PCL significantly outperforms strong actor-critic and Q-learning baselines across several benchmarks.",
"Model-free deep reinforcement learning (RL) algorithms have been demonstrated on a range of challenging decision making and control tasks. However, these methods typically suffer from two major challenges: very high sample complexity and brittle convergence properties, which necessitate meticulous hyperparameter tuning. Both of these challenges severely limit the applicability of such methods to complex, real-world domains. In this paper, we propose soft actor-critic, an off-policy actor-critic deep RL algorithm based on the maximum entropy reinforcement learning framework. In this framework, the actor aims to maximize expected reward while also maximizing entropy - that is, succeed at the task while acting as randomly as possible. Prior deep RL methods based on this framework have been formulated as either off-policy Q-learning, or on-policy policy gradient methods. By combining off-policy updates with a stable stochastic actor-critic formulation, our method achieves state-of-the-art performance on a range of continuous control benchmark tasks, outperforming prior on-policy and off-policy methods. Furthermore, we demonstrate that, in contrast to other off-policy algorithms, our approach is very stable, achieving very similar performance across different random seeds.",
"We propose a method for learning expressive energy-based policies for continuous states and actions, which has been feasible only in tabular domains before. We apply our method to learning maximum entropy policies, resulting into a new algorithm, called soft Q-learning, that expresses the optimal policy via a Boltzmann distribution. We use the recently proposed amortized Stein variational gradient descent to learn a stochastic sampling network that approximates samples from this distribution. The benefits of the proposed algorithm include improved exploration and compositionality that allows transferring skills between tasks, which we confirm in simulated experiments with swimming and walking robots. We also draw a connection to actor-critic methods, which can be viewed performing approximate inference on the corresponding energy-based model."
]
} |
1902.05551 | 2914362802 | We propose a new policy iteration theory as an important extension of soft policy iteration and Soft Actor-Critic (SAC), one of the most efficient model free algorithms for deep reinforcement learning. Supported by the new theory, arbitrary entropy measures that generalize Shannon entropy, such as Tsallis entropy and Renyi entropy, can be utilized to properly randomize action selection while fulfilling the goal of maximizing expected long-term rewards. Our theory gives birth to two new algorithms, i.e., Tsallis entropy Actor-Critic (TAC) and Renyi entropy Actor-Critic (RAC). Theoretical analysis shows that these algorithms can be more effective than SAC. Moreover, they pave the way for us to develop a new Ensemble Actor-Critic (EAC) algorithm in this paper that features the use of a bootstrap mechanism for deep environment exploration as well as a new value-function based mechanism for high-level action selection. Empirically we show that TAC, RAC and EAC can achieve state-of-the-art performance on a range of benchmark control tasks, outperforming SAC and several cutting-edge learning algorithms in terms of both sample efficiency and effectiveness. | The actor-critic framework is commonly utilized by many existing RL algorithms, such as TRPO @cite_26 , PPO @cite_17 @cite_21 and ACKTR @cite_28 . However, a majority of these algorithms must collect a sequence of new environment samples for each learning iteration. Our algorithms, in comparison, can reuse past samples efficiently for policy training. Meanwhile, entropy regularization in the form of KL divergence is often exploited to restrict behavioral changes of updated policies so as to stabilize learning @cite_26 @cite_0 . On the other hand, our algorithms encourage entropy maximization for effective environment exploration. | {
"cite_N": [
"@cite_26",
"@cite_28",
"@cite_21",
"@cite_0",
"@cite_17"
],
"mid": [
"2949608212",
"2963313316",
"2798189842",
"",
"2736601468"
],
"abstract": [
"We describe an iterative procedure for optimizing policies, with guaranteed monotonic improvement. By making several approximations to the theoretically-justified procedure, we develop a practical algorithm, called Trust Region Policy Optimization (TRPO). This algorithm is similar to natural policy gradient methods and is effective for optimizing large nonlinear policies such as neural networks. Our experiments demonstrate its robust performance on a wide variety of tasks: learning simulated robotic swimming, hopping, and walking gaits; and playing Atari games using images of the screen as input. Despite its approximations that deviate from the theory, TRPO tends to give monotonic improvement, with little tuning of hyperparameters.",
"In this work, we propose to apply trust region optimization to deep reinforcement learning using a recently proposed Kronecker-factored approximation to the curvature. We extend the framework of natural policy gradient and propose to optimize both the actor and the critic using Kronecker-factored approximate curvature (K-FAC) with trust region; hence we call our method Actor Critic using Kronecker-Factored Trust Region (ACKTR). To the best of our knowledge, this is the first scalable trust region natural gradient method for actor-critic methods. It is also the method that learns non-trivial tasks in continuous control as well as discrete control policies directly from raw pixel inputs. We tested our approach across discrete domains in Atari games as well as continuous domains in the MuJoCo environment. With the proposed methods, we are able to achieve higher rewards and a 2- to 3-fold improvement in sample efficiency on average, compared to previous state-of-the-art on-policy actor-critic methods. Code is available at https: github.com openai baselines",
"Very recently proximal policy optimization (PPO) algorithms have been proposed as first-order optimization methods for effective reinforcement learning. While PPO is inspired by the same learning theory that justifies trust region policy optimization (TRPO), PPO substantially simplifies algorithm design and improves data efficiency by performing multiple epochs of from sampled data. Although clipping in PPO stands for an important new mechanism for efficient and reliable policy update, it may fail to adaptively improve learning performance in accordance with the importance of each sampled state. To address this issue, a new surrogate learning objective featuring an adaptive clipping mechanism is proposed in this paper, enabling us to develop a new algorithm, known as PPO- @math . PPO- @math optimizes policies repeatedly based on a theoretical target for adaptive policy improvement. Meanwhile, destructively large policy update can be effectively prevented through both clipping and adaptive control of a hyperparameter @math in PPO- @math , ensuring high learning reliability. PPO- @math enjoys the same simple and efficient design as PPO. Empirically on several Atari game playing tasks and benchmark control tasks, PPO- @math also achieved clearly better performance than PPO.",
"",
"We propose a new family of policy gradient methods for reinforcement learning, which alternate between sampling data through interaction with the environment, and optimizing a \"surrogate\" objective function using stochastic gradient ascent. Whereas standard policy gradient methods perform one gradient update per data sample, we propose a novel objective function that enables multiple epochs of minibatch updates. The new methods, which we call proximal policy optimization (PPO), have some of the benefits of trust region policy optimization (TRPO), but they are much simpler to implement, more general, and have better sample complexity (empirically). Our experiments test PPO on a collection of benchmark tasks, including simulated robotic locomotion and Atari game playing, and we show that PPO outperforms other online policy gradient methods, and overall strikes a favorable balance between sample complexity, simplicity, and wall-time."
]
} |
1902.05551 | 2914362802 | We propose a new policy iteration theory as an important extension of soft policy iteration and Soft Actor-Critic (SAC), one of the most efficient model free algorithms for deep reinforcement learning. Supported by the new theory, arbitrary entropy measures that generalize Shannon entropy, such as Tsallis entropy and Renyi entropy, can be utilized to properly randomize action selection while fulfilling the goal of maximizing expected long-term rewards. Our theory gives birth to two new algorithms, i.e., Tsallis entropy Actor-Critic (TAC) and Renyi entropy Actor-Critic (RAC). Theoretical analysis shows that these algorithms can be more effective than SAC. Moreover, they pave the way for us to develop a new Ensemble Actor-Critic (EAC) algorithm in this paper that features the use of a bootstrap mechanism for deep environment exploration as well as a new value-function based mechanism for high-level action selection. Empirically we show that TAC, RAC and EAC can achieve state-of-the-art performance on a range of benchmark control tasks, outperforming SAC and several cutting-edge learning algorithms in terms of both sample efficiency and effectiveness. | Previous works have also studied the maximum entropy framework for both on-policy and off-policy learning @cite_6 @cite_4 @cite_20 . Among them, a large group focused mainly on problems with discrete actions. Some recently developed algorithms have further extended the maximum entropy framework to continuous domains @cite_34 @cite_19 . Different from these algorithms that focus mainly on Shannon entropy, guided by a new policy iteration theory, our newly proposed RL algorithms can maximize general entropy measures such as Tsallis entropy and R 'enyi entropy in continuous action spaces. | {
"cite_N": [
"@cite_4",
"@cite_6",
"@cite_19",
"@cite_34",
"@cite_20"
],
"mid": [
"2554120691",
"2593044849",
"2962902376",
"2949561945",
"2727576081"
],
"abstract": [
"Policy gradient is an efficient technique for improving a policy in a reinforcement learning setting. However, vanilla online variants are on-policy only and not able to take advantage of off-policy data. In this paper we describe a new technique that combines policy gradient with off-policy Q-learning, drawing experience from a replay buffer. This is motivated by making a connection between the fixed points of the regularized policy gradient algorithm and the Q-values. This connection allows us to estimate the Q-values from the action preferences of the policy, to which we apply Q-learning updates. We refer to the new technique as 'PGQL', for policy gradient and Q-learning. We also establish an equivalency between action-value fitting techniques and actor-critic algorithms, showing that regularized policy gradient techniques can be interpreted as advantage function learning algorithms. We conclude with some numerical examples that demonstrate improved data efficiency and stability of PGQL. In particular, we tested PGQL on the full suite of Atari games and achieved performance exceeding that of both asynchronous advantage actor-critic (A3C) and Q-learning.",
"We establish a new connection between value and policy based reinforcement learning (RL) based on a relationship between softmax temporal value consistency and policy optimality under entropy regularization. Specifically, we show that softmax consistent action values correspond to optimal entropy regularized policy probabilities along any action sequence, regardless of provenance. From this observation, we develop a new RL algorithm, Path Consistency Learning (PCL), that minimizes a notion of soft consistency error along multi-step action sequences extracted from both on- and off-policy traces. We examine the behavior of PCL in different scenarios and show that PCL can be interpreted as generalizing both actor-critic and Q-learning algorithms. We subsequently deepen the relationship by showing how a single model can be used to represent both a policy and the corresponding softmax state values, eliminating the need for a separate critic. The experimental evaluation demonstrates that PCL significantly outperforms strong actor-critic and Q-learning baselines across several benchmarks.",
"Model-free deep reinforcement learning (RL) algorithms have been demonstrated on a range of challenging decision making and control tasks. However, these methods typically suffer from two major challenges: very high sample complexity and brittle convergence properties, which necessitate meticulous hyperparameter tuning. Both of these challenges severely limit the applicability of such methods to complex, real-world domains. In this paper, we propose soft actor-critic, an off-policy actor-critic deep RL algorithm based on the maximum entropy reinforcement learning framework. In this framework, the actor aims to maximize expected reward while also maximizing entropy - that is, succeed at the task while acting as randomly as possible. Prior deep RL methods based on this framework have been formulated as either off-policy Q-learning, or on-policy policy gradient methods. By combining off-policy updates with a stable stochastic actor-critic formulation, our method achieves state-of-the-art performance on a range of continuous control benchmark tasks, outperforming prior on-policy and off-policy methods. Furthermore, we demonstrate that, in contrast to other off-policy algorithms, our approach is very stable, achieving very similar performance across different random seeds.",
"We propose a method for learning expressive energy-based policies for continuous states and actions, which has been feasible only in tabular domains before. We apply our method to learning maximum entropy policies, resulting into a new algorithm, called soft Q-learning, that expresses the optimal policy via a Boltzmann distribution. We use the recently proposed amortized Stein variational gradient descent to learn a stochastic sampling network that approximates samples from this distribution. The benefits of the proposed algorithm include improved exploration and compositionality that allows transferring skills between tasks, which we confirm in simulated experiments with swimming and walking robots. We also draw a connection to actor-critic methods, which can be viewed performing approximate inference on the corresponding energy-based model.",
"Trust region methods, such as TRPO, are often used to stabilize policy optimization algorithms in reinforcement learning (RL). While current trust region strategies are effective for continuous control, they typically require a prohibitively large amount of on-policy interaction with the environment. To address this problem, we propose an off-policy trust region method, Trust-PCL. The algorithm is the result of observing that the optimal policy and state values of a maximum reward objective with a relative-entropy regularizer satisfy a set of multi-step pathwise consistencies along any path. Thus, Trust-PCL is able to maintain optimization stability while exploiting off-policy data to improve sample efficiency. When evaluated on a number of continuous control tasks, Trust-PCL improves the solution quality and sample efficiency of TRPO."
]
} |
1902.05551 | 2914362802 | We propose a new policy iteration theory as an important extension of soft policy iteration and Soft Actor-Critic (SAC), one of the most efficient model free algorithms for deep reinforcement learning. Supported by the new theory, arbitrary entropy measures that generalize Shannon entropy, such as Tsallis entropy and Renyi entropy, can be utilized to properly randomize action selection while fulfilling the goal of maximizing expected long-term rewards. Our theory gives birth to two new algorithms, i.e., Tsallis entropy Actor-Critic (TAC) and Renyi entropy Actor-Critic (RAC). Theoretical analysis shows that these algorithms can be more effective than SAC. Moreover, they pave the way for us to develop a new Ensemble Actor-Critic (EAC) algorithm in this paper that features the use of a bootstrap mechanism for deep environment exploration as well as a new value-function based mechanism for high-level action selection. Empirically we show that TAC, RAC and EAC can achieve state-of-the-art performance on a range of benchmark control tasks, outperforming SAC and several cutting-edge learning algorithms in terms of both sample efficiency and effectiveness. | In recent years, ensemble methods have gained wide-spread attention in the research community @cite_38 @cite_33 @cite_13 . For example, Bootstrapped Deep Q-Network and UCB Q-Ensemble @cite_38 @cite_33 have been developed for single-agent RL with success. However, these algorithms were designed to work with discrete actions. In an attempt to tackle continuous problems, some promising methods such as ACE @cite_18 , SOUP @cite_30 and MACE @cite_22 have been further developed. Several relevant algorithms have also been introduced to support ensemble learning under a multi-agent framework @cite_12 @cite_9 . | {
"cite_N": [
"@cite_30",
"@cite_38",
"@cite_18",
"@cite_33",
"@cite_22",
"@cite_9",
"@cite_13",
"@cite_12"
],
"mid": [
"2807908072",
"2963938771",
"2778821583",
"2767313115",
"2460299708",
"2107544712",
"",
"2963407617"
],
"abstract": [
"",
"Efficient exploration remains a major challenge for reinforcement learning (RL). Common dithering strategies for exploration, such as '-greedy, do not carry out temporally-extended (or deep) exploration; this can lead to exponentially larger data requirements. However, most algorithms for statistically efficient RL are not computationally tractable in complex environments. Randomized value functions offer a promising approach to efficient exploration with generalization, but existing algorithms are not compatible with nonlinearly parameterized value functions. As a first step towards addressing such contexts we develop bootstrapped DQN. We demonstrate that bootstrapped DQN can combine deep exploration with deep neural networks for exponentially faster learning than any dithering strategy. In the Arcade Learning Environment bootstrapped DQN substantially improves learning speed and cumulative performance across most games.",
"We introduce an Actor-Critic Ensemble(ACE) method for improving the performance of Deep Deterministic Policy Gradient(DDPG) algorithm. At inference time, our method uses a critic ensemble to select the best action from proposals of multiple actors running in parallel. By having a larger candidate set, our method can avoid actions that have fatal consequences, while staying deterministic. Using ACE, we have won the 2nd place in NIPS'17 Learning to Run competition, under the name of \"Megvii-hzwer\".",
"We show how an ensemble of @math -functions can be leveraged for more effective exploration in deep reinforcement learning. We build on well established algorithms from the bandit setting, and adapt them to the @math -learning setting. We propose an exploration strategy based on upper-confidence bounds (UCB). Our experiments show significant gains on the Atari benchmark.",
"Reinforcement learning offers a promising methodology for developing skills for simulated characters, but typically requires working with sparse hand-crafted features. Building on recent progress in deep reinforcement learning (DeepRL), we introduce a mixture of actor-critic experts (MACE) approach that learns terrain-adaptive dynamic locomotion skills using high-dimensional state and terrain descriptions as input, and parameterized leaps or steps as output actions. MACE learns more quickly than a single actor-critic approach and results in actor-critic experts that exhibit specialization. Additional elements of our solution that contribute towards efficient learning include Boltzmann exploration and the use of initial actor biases to encourage specialization. Results are demonstrated for multiple planar characters and terrain classes.",
"Cooperative multi-agent systems (MAS) are ones in which several agents attempt, through their interaction, to jointly solve tasks or to maximize utility. Due to the interactions among the agents, multi-agent problem complexity can rise rapidly with the number of agents or their behavioral sophistication. The challenge this presents to the task of programming solutions to MAS problems has spawned increasing interest in machine learning techniques to automate the search and optimization process. We provide a broad survey of the cooperative multi-agent learning literature. Previous surveys of this area have largely focused on issues common to specific subareas (for example, reinforcement learning, RL or robotics). In this survey we attempt to draw from multi-agent learning work in a spectrum of areas, including RL, evolutionary computation, game theory, complex systems, agent modeling, and robotics. We find that this broad view leads to a division of the work into two categories, each with its own special issues: applying a single learner to discover joint solutions to multi-agent problems (team learning), or using multiple simultaneous learners, often one per agent (concurrent learning). Additionally, we discuss direct and indirect communication in connection with learning, plus open issues in task decomposition, scalability, and adaptive dynamics. We conclude with a presentation of multi-agent learning problem domains, and a list of multi-agent learning resources.",
"",
"We explore deep reinforcement learning methods for multi-agent domains. We begin by analyzing the difficulty of traditional algorithms in the multi-agent case: Q-learning is challenged by an inherent non-stationarity of the environment, while policy gradient suffers from a variance that increases as the number of agents grows. We then present an adaptation of actor-critic methods that considers action policies of other agents and is able to successfully learn policies that require complex multi-agent coordination. Additionally, we introduce a training regimen utilizing an ensemble of policies for each agent that leads to more robust multi-agent policies. We show the strength of our approach compared to existing methods in cooperative as well as competitive scenarios, where agent populations are able to discover various physical and informational coordination strategies."
]
} |
1902.05378 | 2892302299 | Selecting an optimal set of icons is a crucial step in the pipeline of visual design to structure and navigate through content. However, designing the icons sets is usually a difficult task for which expert knowledge is required. In this work, to ease the process of icon set selection to the users, we propose a similarity metric which captures the properties of style and visual identity. We train a Siamese Neural Network with an on-line dataset of icons organized in visually coherent collections that are used to adaptively sample training data and optimize the training process. As the dataset contains noise, we further collect human-rated information on the perception of icon’s similarity which will be used for evaluating and testing the proposed model. We present several results and applications based on searches, kernel visualizations and optimized set proposals that can be helpful for designers and non-expert users while exploring large collections of icons. | Previous works have focused on generating semantically relevant icons to improve visualizations @cite_15 @cite_30 . In particular, Setlur and Mackinlay @cite_30 develop a method for mapping categorical data to icons. They found out that users prefer stylistically similar icons within a set, as opposed to automatic sets that might differ in look-and-feel. @cite_45 studied how the perception of icons is affected by spatial layouts, and present a shape grammar to generate visually distinctive icons. Our work is inspired by these, although we propose a deep learning-based method to measure style and visual identity between icons. | {
"cite_N": [
"@cite_30",
"@cite_15",
"@cite_45"
],
"mid": [
"2057592763",
"2168635901",
"1986349338"
],
"abstract": [
"Authors use icon encodings to indicate the semantics of categorical information in visualizations. The default icon libraries found in visualization tools often do not match the semantics of the data. Users often manually search for or create icons that are more semantically meaningful. This process can hinder the flow of visual analysis, especially when the amount of data is large, leading to a suboptimal user experience. We propose a technique for automatically generating semantically relevant icon encodings for categorical dimensions of data points. The algorithm employs natural language processing in order to find relevant imagery from the Internet. We evaluate our approach on Mechanical Turk by generating large libraries of icons using Tableau Public workbooks that represent real analytical effort by people out in the world. Our results show that the automatic algorithm does nearly as well as the manually created icons, and particularly has higher user satisfaction for larger cardinalities of data.",
"Semanticons can enhance the representation of files by offering symbols that are both meaningful and easily distinguishable. The semantics of a file is estimated by parsing its name, location, and content to generate a ‘context’, which is used to query an image database. The resulting images are simplified by segmenting them, computing an importance value for each segmented region, and removing unimportant regions. The",
"Although existing GUIs have a sense of space, they provide no sense of place. Numerous studies report that users misplace files and have trouble wayfinding in virtual worlds despite the fact that people have remarkable visual and spatial abilities. This issue is considered in the human-computer interface field and has been addressed with alternate display navigation schemes. Our paper presents a fundamentally graphics based approach to this 'lost in hyperspace' problem. Specifically, we propose that spatial display of files is not sufficient to engage our visual skills; scenery (distinctive visual appearance) is needed as well. While scenery (in the form of custom icon assignments) is already possible in current operating systems, few if any users take the time to manually assign icons to all their files. As such, our proposal is to generate visually distinctive icons (\"VisualIDs\") automatically, while allowing the user to replace the icon if desired. The paper discusses psychological and conceptual issues relating to icons, visual memory, and the necessary relation of scenery to data. A particular icon generation algorithm is described; subjects using these icons in simulated file search and recall tasks show significantly improved performance with little effort. Although the incorporation of scenery in a graphical user interface will introduce many new (and interesting) design problems that cannot be addressed in this paper, we show that automatically created scenery is both beneficial and feasible."
]
} |
1902.05378 | 2892302299 | Selecting an optimal set of icons is a crucial step in the pipeline of visual design to structure and navigate through content. However, designing the icons sets is usually a difficult task for which expert knowledge is required. In this work, to ease the process of icon set selection to the users, we propose a similarity metric which captures the properties of style and visual identity. We train a Siamese Neural Network with an on-line dataset of icons organized in visually coherent collections that are used to adaptively sample training data and optimize the training process. As the dataset contains noise, we further collect human-rated information on the perception of icon’s similarity which will be used for evaluating and testing the proposed model. We present several results and applications based on searches, kernel visualizations and optimized set proposals that can be helpful for designers and non-expert users while exploring large collections of icons. | Style similarity metrics have been recently proposed for fonts @cite_2 , infographics @cite_55 , 3D models @cite_28 @cite_52 , or interior designs @cite_8 . Closer to our goal, the work of @cite_33 uses a hand-made feature vector to measure style similarity for clip art. However, since the feature descriptors were manually selected for that particular task, and do not account for high-level properties, their distance metric does not generalize to our data, as we will show later. In a follow-up work, @cite_36 find that shape is a property that people take into account when comparing clip arts, however, it is not measured in their existing style metric for clip art. On the contrary, we automatically learn a distance metric that measures both style and visual identity using a deep Siamese Neural Network trained from scratch. | {
"cite_N": [
"@cite_33",
"@cite_8",
"@cite_28",
"@cite_36",
"@cite_55",
"@cite_52",
"@cite_2"
],
"mid": [
"1983764898",
"2021354639",
"2064723547",
"2466313891",
"2964340947",
"2090621143",
"1981219830"
],
"abstract": [
"This paper presents a method for measuring the similarity in style between two pieces of vector art, independent of content. Similarity is measured by the differences between four types of features: color, shading, texture, and stroke. Feature weightings are learned from crowdsourced experiments. This perceptual similarity enables style-based search. Using our style-based search feature, we demonstrate an application that allows users to create stylistically-coherent clip art mash-ups.",
"Popular sites like Houzz, Pinterest, and LikeThatDecor, have communities of users helping each other answer questions about products in images. In this paper we learn an embedding for visual search in interior design. Our embedding contains two different domains of product images: products cropped from internet scenes, and products in their iconic form. With such a multi-domain embedding, we demonstrate several applications of visual search including identifying products in scenes and finding stylistically similar products. To obtain the embedding, we train a convolutional neural network on pairs of images. We explore several training architectures including re-purposing object classifiers, using siamese networks, and using multitask learning. We evaluate our search quantitatively and qualitatively and demonstrate high quality results for search across multiple visual domains, enabling new applications in interior design.",
"The human perception of stylistic similarity transcends structure and function: for instance, a bed and a dresser may share a common style. An algorithmically computed style similarity measure that mimics human perception can benefit a range of computer graphics applications. Previous work in style analysis focused on shapes within the same class, and leveraged structural similarity between these shapes to facilitate analysis. In contrast, we introduce the first structure-transcending style similarity measure and validate it to be well aligned with human perception of stylistic similarity. Our measure is inspired by observations about style similarity in art history literature, which point to the presence of similarly shaped, salient, geometric elements as one of the key indicators of stylistic similarity. We translate these observations into an algorithmic measure by first quantifying the geometric properties that make humans perceive geometric elements as similarly shaped and salient in the context of style, then employing this quantification to detect pairs of matching style related elements on the analyzed models, and finally collating the element-level geometric similarity measurements into an object-level style measure consistent with human perception. To achieve this consistency we employ crowdsourcing to quantify the different components of our measure; we learn the relative perceptual importance of a range of elementary shape distances and other parameters used in our measurement from 50K responses to cross-structure style similarity queries provided by over 2500 participants.We train and validate our method on this dataset, showing it to successfully predict relative style similarity with near 90 accuracy based on 10-fold cross-validation.",
"Searching by style in illustration data sets is a particular problem in Information Retrieval which has received little attention so far. One of its main problems is that the perception of style is highly subjective, which makes labeling styles a very difficult task. Despite being difficult to predict computationally, certain properties such as colorfulness, line style or shading can be successfully captured by existing style metrics. However, there is little knowledge about how we distinguish between different styles and how these metrics can be used to guide users in style-based interactions. In this paper, we propose several contributions towards a better comprehension of illustration style and its usefulness for data exploration and retrieval. First, we provide new insights about how we perceive style in illustration. Second, we evaluate a handmade style clustering of clip art pieces with an existing style metric to analyze how this metric aligns with expert knowledge. Finally, we propose a method for efficient navigation and exploration of large clip art data sets which takes into account both semantic labeling of the data and its style. Our approach combines hierarchical clustering with dimensionality reduction techniques, and strategic sampling to obtain intuitive visualizations and useful visualizations.",
"Infographics are complex graphic designs integrating text, images, charts and sketches. Despite the increasing popularity of infographics and the rapid growth of online design portfolios, little research investigates how we can take advantage of these design resources. In this paper we present a method for measuring the style similarity between infographics. Based on human perception data collected from crowdsourced experiments, we use computer vision and machine learning algorithms to learn a style similarity metric for infographic designs. We evaluate different visual features and learning algorithms and find that a combination of color histograms and Histograms-of-Gradients (HoG) features is most effective in characterizing the style of infographics. We demonstrate our similarity metric on a preliminary image retrieval test.",
"This paper presents a method for learning to predict the stylistic compatibility between 3D furniture models from different object classes: e.g., how well does this chair go with that table? To do this, we collect relative assessments of style compatibility using crowdsourcing. We then compute geometric features for each 3D model and learn a mapping of them into a space where Euclidean distances represent style incompatibility. Motivated by the geometric subtleties of style, we introduce part-aware geometric feature vectors that characterize the shapes of different parts of an object separately. Motivated by the need to compute style compatibility between different object classes, we introduce a method to learn object class-specific mappings from geometric features to a shared feature space. During experiments with these methods, we find that they are effective at predicting style compatibility agreed upon by people. We find in user studies that the learned compatibility metric is useful for novel interactive tools that: 1) retrieve stylistically compatible models for a query, 2) suggest a piece of furniture for an existing scene, and 3) help guide an interactive 3D modeler towards scenes with compatible furniture.",
"This paper presents interfaces for exploring large collections of fonts for design tasks. Existing interfaces typically list fonts in a long, alphabetically-sorted menu that can be challenging and frustrating to explore. We instead propose three interfaces for font selection. First, we organize fonts using high-level descriptive attributes, such as \"dramatic\" or \"legible.\" Second, we organize fonts in a tree-based hierarchical menu based on perceptual similarity. Third, we display fonts that are most similar to a user's currently-selected font. These tools are complementary; a user may search for \"graceful\" fonts, select a reasonable one, and then refine the results from a list of fonts similar to the selection. To enable these tools, we use crowdsourcing to gather font attribute data, and then train models to predict attribute values for new fonts. We use attributes to help learn a font similarity metric using crowdsourced comparisons. We evaluate the interfaces against a conventional list interface and find that our interfaces are preferred to the baseline. Our interfaces also produce better results in two real-world tasks: finding the nearest match to a target font, and font selection for graphic designs."
]
} |
1902.05378 | 2892302299 | Selecting an optimal set of icons is a crucial step in the pipeline of visual design to structure and navigate through content. However, designing the icons sets is usually a difficult task for which expert knowledge is required. In this work, to ease the process of icon set selection to the users, we propose a similarity metric which captures the properties of style and visual identity. We train a Siamese Neural Network with an on-line dataset of icons organized in visually coherent collections that are used to adaptively sample training data and optimize the training process. As the dataset contains noise, we further collect human-rated information on the perception of icon’s similarity which will be used for evaluating and testing the proposed model. We present several results and applications based on searches, kernel visualizations and optimized set proposals that can be helpful for designers and non-expert users while exploring large collections of icons. | To measure shape similarity is a long-standing problem in computer graphics with many different approaches trying to solve it. Bober @cite_4 shows how to represent and match shape representations under the MPEG-7 standard @cite_16 . @cite_26 propose several silhouette-based descriptors that can be used for 2D and 3D shape retrieval. Other shape descriptors have been proposed, including Hu-moments @cite_35 , shape context @cite_19 , the use of Zernike moments @cite_12 , pyramid of arclength descriptors @cite_9 , or Fourier descriptors @cite_17 . @cite_47 focused on 3D shape similarity, using part-based models, while other works compare shapes using single closed contours @cite_20 @cite_21 . In contrast, our method does not need to explicitly model the geometrical properties of the given image and implicitly considers additional properties such as image abstraction and complexity that are recognized while training the Siamese Neural Network. | {
"cite_N": [
"@cite_35",
"@cite_26",
"@cite_4",
"@cite_9",
"@cite_21",
"@cite_19",
"@cite_47",
"@cite_16",
"@cite_20",
"@cite_12",
"@cite_17"
],
"mid": [
"2159498975",
"",
"2049092359",
"2553667411",
"2105026917",
"2057175746",
"2012604613",
"2105847589",
"2106404777",
"2118376687",
"2027393660"
],
"abstract": [
"In this paper a theory of two-dimensional moment invariants for planar geometric figures is presented. A fundamental theorem is established to relate such moment invariants to the well-known algebraic invariants. Complete systems of moment invariants under translation, similitude and orthogonal transformations are derived. Some moment invariants under general two-dimensional linear transformations are also included. Both theoretical formulation and practical models of visual pattern recognition based upon these moment invariants are discussed. A simple simulation program together with its performance are also presented. It is shown that recognition of geometrical patterns and alphabetical characters independently of position, size and orientation can be accomplished. It is also indicated that generalization is possible to include invariance with parallel projection.",
"",
"We present a robust and efficient approach to video stabilization that achieves high-quality camera motion for a wide range of videos. In this article, we focus on the problem of transforming a set of input 2D motion trajectories so that they are both smooth and resemble visually plausible views of the imaged scene; our key insight is that we can achieve this goal by enforcing subspace constraints on feature trajectories while smoothing them. Our approach assembles tracked features in the video into a trajectory matrix, factors it into two low-rank matrices, and performs filtering or curve fitting in a low-dimensional linear space. In order to process long videos, we propose a moving factorization that is both efficient and streamable. Our experiments confirm that our approach can efficiently provide stabilization results comparable with prior 3D methods in cases where those methods succeed, but also provides smooth camera motions in cases where such approaches often fail, such as videos that lack parallax. The presented approach offers the first method that both achieves high-quality video stabilization and is practical enough for consumer applications.",
"This paper tackles a challenging 2D collage generation problem, focusing on shapes: we aim to fill a given region by packing irregular and reasonably-sized shapes with minimized gaps and overlaps. To achieve this nontrivial problem, we first have to analyze the boundary of individual shapes and then couple the shapes with partially-matched boundary to reduce gaps and overlaps in the collages. Second, the search space in identifying a good coupling of shapes is highly enormous, since arranging a shape in a collage involves a position, an orientation, and a scale factor. Yet, this matching step needs to be performed for every single shape when we pack it into a collage. Existing shape descriptors are simply infeasible for computation in a reasonable amount of time. To overcome this, we present a brand new, scale- and rotation-invariant 2D shape descriptor, namely pyramid of arclength descriptor (PAD). Its formulation is locally supported, scalable, and yet simple to construct and compute. These properties make PAD efficient for performing the partial-shape matching. Hence, we can prune away most search space with simple calculation, and efficiently identify candidate shapes. We evaluate our method using a large variety of shapes with different types and contours. Convincing collage results in terms of visual quality and time performance are obtained.",
"Shape similarity and shape retrieval are very important topics in computer vision. The recent progress in this domain has been mostly driven by designing smart shape descriptors for providing better similarity measure between pairs of shapes. In this paper, we provide a new perspective to this problem by considering the existing shapes as a group, and study their similarity measures to the query shape in a graph structure. Our method is general and can be built on top of any existing shape similarity measure. For a given similarity measure, a new similarity is learned through graph transduction. The new similarity is learned iteratively so that the neighbors of a given shape influence its final similarity to the query. The basic idea here is related to PageRank ranking, which forms a foundation of Google Web search. The presented experimental results demonstrate that the proposed approach yields significant improvements over the state-of-art shape matching algorithms. We obtained a retrieval rate of 91.61 percent on the MPEG-7 data set, which is the highest ever reported in the literature. Moreover, the learned similarity by the proposed method also achieves promising improvements on both shape classification and shape clustering.",
"We present a novel approach to measuring similarity between shapes and exploit it for object recognition. In our framework, the measurement of similarity is preceded by: (1) solving for correspondences between points on the two shapes; (2) using the correspondences to estimate an aligning transform. In order to solve the correspondence problem, we attach a descriptor, the shape context, to each point. The shape context at a reference point captures the distribution of the remaining points relative to it, thus offering a globally discriminative characterization. Corresponding points on two similar shapes will have similar shape contexts, enabling us to solve for correspondences as an optimal assignment problem. Given the point correspondences, we estimate the transformation that best aligns the two shapes; regularized thin-plate splines provide a flexible class of transformation maps for this purpose. The dissimilarity between the two shapes is computed as a sum of matching errors between corresponding points, together with a term measuring the magnitude of the aligning transform. We treat recognition in a nearest-neighbor classification framework as the problem of finding the stored prototype shape that is maximally similar to that in the image. Results are presented for silhouettes, trademarks, handwritten digits, and the COIL data set.",
"Computing similarities or distances between 3D shapes is a crucial building block for numerous tasks, including shape retrieval, exploration and classification. Current state-of-the-art distance measures mostly consider the overall appearance of the shapes and are less sensitive to fine changes in shape structure or geometry. We present shape edit distance (SHED) that measures the amount of effort needed to transform one shape into the other, in terms of re-arranging the parts of one shape to match the parts of the other shape, as well as possibly adding and removing parts. The shape edit distance takes into account both the similarity of the overall shape structure and the similarity of individual parts of the shapes. We show that SHED is favorable to state-of-the-art distance measures in a variety of applications and datasets, and is especially successful in scenarios where detecting fine details of the shapes is important, such as shape retrieval and exploration.",
"The MPEG-7 visual standard under development specifies content-based descriptors that allow users or agents (or search engines) to measure similarity in images or video based on visual criteria, and can be used to efficiently identify, filter, or browse images or video based on visual content. More specifically, MPEG-7 specifies color, texture, object shape, global motion, or object motion features for this purpose. This paper outlines the aim, methodologies, and broad details of the MPEG-7 standard development for visual content description.",
"The Core Experiment CE-Shape-1 for shape descriptors performed for the MPEG-7 standard gave a unique opportunity to compare various shape descriptors for non-rigid shapes with a single closed contour. There are two main differences with respect to other comparison results reported in the literature: (1) For each shape descriptor the experiments were carried out by an institute that is in favor of this descriptor. This implies that the parameters for each system were optimally determined and the implementations were thoroughly rested. (2) It was possible to compare the performance of shape descriptors based on totally different mathematical approaches. A more theoretical comparison of these descriptors seems to be extremely hard. In this paper we report on the MPEG-7 Core Experiment CE-Shape.",
"The problem of rotation-, scale-, and translation-invariant recognition of images is discussed. A set of rotation-invariant features are introduced. They are the magnitudes of a set of orthogonal complex moments of the image known as Zernike moments. Scale and translation invariance are obtained by first normalizing the image with respect to these parameters using its regular geometrical moments. A systematic reconstruction-based method for deciding the highest-order Zernike moments required in a classification problem is developed. The quality of the reconstructed image is examined through its comparison to the original one. The orthogonality property of the Zernike moments, which simplifies the process of image reconstruction, make the suggest feature selection approach practical. Features of each order can also be weighted according to their contribution to the reconstruction process. The superiority of Zernike moment features over regular moments and moment invariants was experimentally verified. >",
"Shape description is one of the key parts of image content description for image retrieval. Most of the existing shape descriptors are usually either application dependent or non-robust, making them undesirable for generic shape description. In this paper, a generic Fourier descriptor (GFD) is proposed to overcome the drawbacks of existing shape representation techniques. The proposed shape descriptor is derived by applying two-dimensional Fourier transform on a polar-raster sampled shape image. The acquired shape descriptor is application independent and robust. Experimental results show that the proposed GFD outperforms common contour-based and region-based shape descriptors."
]
} |
1902.05260 | 2912338332 | Offchain networks emerge as a promising solution to address the scalability challenge of blockchain. Participants directly make payments through a network of payment channels without the overhead of committing onchain transactions. Routing is critical to the performance of offchain networks. Existing solutions use either static routing with poor performance or dynamic routing with high overhead to obtain the dynamic channel balance information. In this paper, we propose Flash, a new dynamic routing solution that leverages the unique characteristics of transactions in offchain networks to strike a better tradeoff between path optimality and probing overhead. By studying the traces of real offchain networks, we find that the payment sizes are heavy-tailed, and most payments are highly recurrent. Flash thus differentiates the treatment of elephant payments from mice payments. It uses a modified max-flow algorithm for elephant payments to find paths with sufficient capacity, and strategically routes the payment across paths to minimize the transaction fees. Mice payments are directly sent by looking up a routing table with a few precomputed paths to reduce probing overhead. Testbed experiments and data-driven simulations show that Flash improves the success volume of payments by up to 2.3x compared to the state-of-the-art routing algorithm. | The above routing algorithms fall into static routing, which does not consider payment channel dynamics and leads to poor throughput performance. Revive @cite_24 and Spider @cite_14 take the dynamic channel balances into consideration and propose centralized offline routing algorithms to maximize the throughput or success volume of payments. As we discussed centralized schemes have high probing overhead and do not work for decentralized offchain networks. | {
"cite_N": [
"@cite_24",
"@cite_14"
],
"mid": [
"2766452458",
"2890600083"
],
"abstract": [
"Scaling the transaction throughput of decentralized blockchain ledgers such as Bitcoin and Ethereum has been an ongoing challenge. Two-party duplex payment channels have been designed and used as building blocks to construct linked payment networks, which allow atomic and trust-free payments between parties without exhausting the resources of the blockchain. Once a payment channel, however, is depleted (e.g., because transactions were mostly unidirectional) the channel would need to be closed and re-funded to allow for new transactions. Users are envisioned to entertain multiple payment channels with different entities, and as such, instead of refunding a channel (which incurs costly on-chain transactions), a user should be able to leverage his existing channels to rebalance a poorly funded channel. To the best of our knowledge, we present the first solution that allows an arbitrary set of users in a payment channel network to securely rebalance their channels, according to the preferences of the channel owners. Except in the case of disputes (similar to conventional payment channels), our solution does not require on-chain transactions and therefore increases the scalability of existing blockchains. In our security analysis, we show that an honest participant cannot lose any of its funds while rebalancing. We finally provide a proof of concept implementation and evaluation for the Ethereum network.",
"With the growing usage of Bitcoin and other cryptocurrencies, many scalability challenges have emerged. A promising scaling solution, exemplified by the Lightning Network, uses a network of bidirectional payment channels that allows fast transactions between two parties. However, routing payments on these networks efficiently is non-trivial, since payments require finding paths with sufficient funds, and channels can become unidirectional over time blocking further transactions through them. Today's payment channel networks exacerbate these problems by attempting to deliver all payments atomically. In this paper, we present the Spider network, a new packet-switched architecture for payment channel networks. Spider splits payments into transaction units and transmits them over time across different paths. Spider uses congestion control, payment scheduling, and imbalance-aware routing to optimize delivery of payments. Our results show that Spider improves the volume and number of successful payments on the network by 10-45 and 5-40 respectively compared to state-of-the-art approaches."
]
} |
1902.05234 | 2913854095 | It has been widely accepted that Graphics Processing Units (GPU) is one of promising schemes for encryption acceleration, in particular, the support of complex mathematical calculations such as integer and logical operations makes the implementation easier; however, complexes such as parallel granularity, memory allocation still imposes a burden on real world implementations. In this paper, we propose a new approach for Advanced Encryption Standard accelerations, including both encryption and decryption. Specifically, we adapt the Electronic Code Book mode for cryptographic transformation, look up table scheme for fast lookup, and a granularity of one state per thread for thread scheduling. Our experimental results offer researchers a good understanding on GPU architectures and software accelerations. In addition, both our source code and experimental results are freely available. | , @cite_9 propose an implementation of the AES-128 ECB Encryption on three different GPU architectures (Kepler, Maxwell and Pascal). The results show that encryption speeds with 207 Gbps on the NVIDIA GTX TITAN X (Maxwell) and 280 Gbps on the NVIDIA GTX 1080 (Pascal) have been achieved by performing new optimization techniques using 32bytes thread granularity. | {
"cite_N": [
"@cite_9"
],
"mid": [
"2783704790"
],
"abstract": [
"The importance of cryptography on ensuring security or integrity of the electronic data transaction had become higher during the past few years. Multiple security protocols are currently using various block ciphers. One of the most widely used block ciphers is the Advanced Encryption Standard (AES) which is chosen as a standard for its higher efficiency and stronger security than its competitors. Unfortunately, the encryption and decryption processes of AES takes a considerable amount of time for large data size. The GPU is an attractive platform for accelerating block ciphers and other cryptography algorithms due to its massively parallel processing power. In this work, an implementation of the AES-128 ECB Encryption on three different GPU architectures (Kepler, Maxwell and Pascal) has been presented. The results show that encryption speeds with 207 Gbps on the NVIDIA GTX TITAN X (Maxwell) and 280 Gbps on the NVIDIA GTX 1080 (Pascal) have been achieved by performing new optimization techniques using 32bytes thread granularity."
]
} |
1902.05234 | 2913854095 | It has been widely accepted that Graphics Processing Units (GPU) is one of promising schemes for encryption acceleration, in particular, the support of complex mathematical calculations such as integer and logical operations makes the implementation easier; however, complexes such as parallel granularity, memory allocation still imposes a burden on real world implementations. In this paper, we propose a new approach for Advanced Encryption Standard accelerations, including both encryption and decryption. Specifically, we adapt the Electronic Code Book mode for cryptographic transformation, look up table scheme for fast lookup, and a granularity of one state per thread for thread scheduling. Our experimental results offer researchers a good understanding on GPU architectures and software accelerations. In addition, both our source code and experimental results are freely available. | , @cite_18 develope G-BLASTN, a GPU-accelerated nucleotide alignment tool based on the widely used NCBI-BLAST. G-BLASTN can produce exactly the same results as NCBI-BLAST, and it has very similar user commands. Compared with the sequential NCBI-BLAST, G-BLASTN can achieve an overall speedup of 14.80X under ‘megablast’ mode. And they @cite_1 propose to exploit the computing power of Graphic Processing Units (GPUs) for homomorphic hashing. Specifically, they demonstrate how to use NVIDIA GPUs and the Computer Unified Device Architecture (CUDA) programming model to achieve 38 times of speedup over the CPU counterpart. They also develop a multi-precision modular arithmetic library on CUDA platform, which is not only key to our specific application, but also very useful for a large number of cryptographic applications. Also, they @cite_5 propose a GPU based AES implementation of which the frequently accessed T-boxes were allocated on on-chip shared memory and the granularity that one thread handles a 16 bytes AES block was adopted. They achieved the performance of around 60 Gbps throughput on NVIDIA Tesla C2050 GPU, which runs up to 50 times faster than a sequential implementation based on Intel Core i7-920 2.66GHz CPU. | {
"cite_N": [
"@cite_5",
"@cite_18",
"@cite_1"
],
"mid": [
"1965357124",
"2163932369",
"1975244381"
],
"abstract": [
"GPU is continuing its trend of vastly outperforming CPU while becoming more general purpose. In order to improve the efficiency of AES algorithm, this paper proposed a CUDA implementation of Electronic Codebook (ECB) mode encoding process and Cipher Feedback (CBC) mode decoding process on GPU. In our implementation, the frequently accessed T-boxes were allocated on on-chip shared memory and the granularity that one thread handles a 16 Bytes AES block was adopted. Finally, we achieved the highest performance of around 60 Gbps throughput on NVIDIA Tesla C2050 GPU, which runs up to 50 times faster than a sequential implementation based on Intel Core i7-920 2.66GHz CPU. In addition, we discussed the optimization under some practical application scenarios such as overlapping GPU processing and data transfer.",
"Motivation: Since 1990, the basic local alignment search tool (BLAST) has become one of the most popular and fundamental bioinformatics tools for sequence similarity searching, receiving extensive attention from the research community. The two pioneering papers on BLAST have received over 96 000 citations. Given the huge population of BLAST users and the increasing size of sequence databases, an urgent topic of study is how to improve the speed. Recently, graphics processing units (GPUs) have been widely used as low-cost, highperformance computing platforms. The existing GPU-BLAST is a promising software tool that uses a GPU to accelerate protein sequence alignment. Unfortunately, there is still no GPU-accelerated software tool for BLAST-based nucleotide sequence alignment. Results: We developed G-BLASTN, a GPU-accelerated nucleotide alignment tool based on the widely used NCBI-BLAST. G-BLASTN can produce exactly the same results as NCBI-BLAST, and it has very similar user commands. Compared with the sequential NCBIBLAST, G-BLASTN can achieve an overall speedup of 14.80X under ‘megablast’ mode. More impressively, it achieves an overall speedup of 7.15X over the multithreaded NCBI-BLAST running on 4 CPU cores. When running under ‘blastn’ mode, the overall speedups are 4.32X (against 1-core) and 1.56X (against 4-core). G-BLASTN also supports a pipeline mode that further improves the overall performance by up to 44 when handling a batch of queries as a whole. Currently G-BLASTN is best optimized for databases with long sequences. We plan to optimize its performance on short database sequences",
"Multiple-precision integer operations are key components of many security applications; but unfortunately they are computationally expensive on contemporary CPUs. In this paper, we present our design and implementation of a multiple-precision integer library for GPUs which is implemented by CUDA. We report our experimental results which show that a significant speedup can be achieved by GPUs as compared with the GNU MP library on CPUs."
]
} |
1902.05234 | 2913854095 | It has been widely accepted that Graphics Processing Units (GPU) is one of promising schemes for encryption acceleration, in particular, the support of complex mathematical calculations such as integer and logical operations makes the implementation easier; however, complexes such as parallel granularity, memory allocation still imposes a burden on real world implementations. In this paper, we propose a new approach for Advanced Encryption Standard accelerations, including both encryption and decryption. Specifically, we adapt the Electronic Code Book mode for cryptographic transformation, look up table scheme for fast lookup, and a granularity of one state per thread for thread scheduling. Our experimental results offer researchers a good understanding on GPU architectures and software accelerations. In addition, both our source code and experimental results are freely available. | , @cite_14 present results of several experiments that were conducted to elucidate the relation between memory allocation styles of variables of AES and granularity as the parallelism exploited from AES encoding processes using CUDA with an NVIDIA GeForce GTX285 (Nvidia Corp.). Results of these experiments showed that the 16 bytes thread granularity had the highest performance. It achieved approximately 35 Gbps throughput. | {
"cite_N": [
"@cite_14"
],
"mid": [
"1938381964"
],
"abstract": [
"GPU exhibits the capability for applications with a high level of parallelism despite its low cost. The support of integer and logical instructions by the latest generation of GPUs enables us to implement cipher algorithms more easily. However, decisions such as parallel processing granularity and memory allocation impose a heavy burden on programmers. Therefore, this paper presents results of several experiments that were conducted to elucidate the relation between memory allocation styles of variables of AES and granularity as the parallelism exploited from AES encoding processes using CUDA with an NVIDIA GeForce GTX285 (Nvidia Corp.). Results of these experiments showed that the 16 bytes thread granularity had the highest performance. It achieved approximately 35 Gbps throughput. It also exhibited differences of memory allocation and granularity effects around 2 –30 for performance in standard implementation. It shows that the decision of granularity and memory allocation is the most important factor for effective processing in AES encryption on GPU. Moreover, implementation with overlapping between processing and data transfer yielded 22.5 Gbps throughput including the data transfer time."
]
} |
1902.05234 | 2913854095 | It has been widely accepted that Graphics Processing Units (GPU) is one of promising schemes for encryption acceleration, in particular, the support of complex mathematical calculations such as integer and logical operations makes the implementation easier; however, complexes such as parallel granularity, memory allocation still imposes a burden on real world implementations. In this paper, we propose a new approach for Advanced Encryption Standard accelerations, including both encryption and decryption. Specifically, we adapt the Electronic Code Book mode for cryptographic transformation, look up table scheme for fast lookup, and a granularity of one state per thread for thread scheduling. Our experimental results offer researchers a good understanding on GPU architectures and software accelerations. In addition, both our source code and experimental results are freely available. | , @cite_8 propose an AES-128 algorithm (ECB mode) implementation on three different GPU architectures with different values of granularities (32,64 and 128 bytes thread). Their results show that the throughput factor reaches 277 Gbps, 201 Gbps and 78 Gbps using the NVIDIA GTX 1080 (Pascal), the NVIDIA GTX TITAN X (Maxwell) and the GTX 780 (Kepler) GPU architectures. | {
"cite_N": [
"@cite_8"
],
"mid": [
"2757370574"
],
"abstract": [
"The Advanced Encryption Standard (AES) is One of the most popular symmetric block cipher because it has better efficiency and security. The AES is computation intensive algorithm especially for massive transactions. The Graphics Processing Unit (GPU) is an amazing platform for accelerating AES. it has good parallel processing power. Traditional approaches for implementing AES using GPU use 16 byte per thread as a default granularity. In this paper, the AES-128 algorithm (ECB mode) is implemented on three different GPU architectures with different values of granularities (32,64 and 128 bytes thread). Our results show that the throughput factor reaches 277 Gbps, 201 Gbps and 78 Gbps using the NVIDIA GTX 1080 (Pascal), the NVIDIA GTX TITAN X (Maxwell) and the GTX 780 (Kepler) GPU architectures."
]
} |
1902.05234 | 2913854095 | It has been widely accepted that Graphics Processing Units (GPU) is one of promising schemes for encryption acceleration, in particular, the support of complex mathematical calculations such as integer and logical operations makes the implementation easier; however, complexes such as parallel granularity, memory allocation still imposes a burden on real world implementations. In this paper, we propose a new approach for Advanced Encryption Standard accelerations, including both encryption and decryption. Specifically, we adapt the Electronic Code Book mode for cryptographic transformation, look up table scheme for fast lookup, and a granularity of one state per thread for thread scheduling. Our experimental results offer researchers a good understanding on GPU architectures and software accelerations. In addition, both our source code and experimental results are freely available. | , @cite_6 implement AES decryption in CBC mode, a mode that is widely used by many applications, on GPU using Cuda, a framework developed by NVIDIA and friendly to use. To achieve the best performance, they give a comprehensive performance analysis to the implementations based on GPU under different parameters setting that include the size of input data, the number of threads per block, memory allocation style and parallel granularity. Through experiment evaluation based on our implementation, the best performance of AES on GPU is 112 times over the serial AES algorithm on CPU. | {
"cite_N": [
"@cite_6"
],
"mid": [
"2744618991"
],
"abstract": [
"The Advanced Encryption Standard(AES) is used in security areas widely now. However, there is still a large room for further improvement of its execution efficiency. Since the graphics processing unit(GPU) with potent ability of parallel computing has been applied in general purpose of computation, people have tried to use it to faster execution time in various cryptographic algorithms. This paper discusses how the performance of CBCAES decryption based on GPU is influenced by 4 key parameters that include the size of input data, the number of threads per block, memory allocation style and parallel granularity. Further more, we compare the performance of AES on GPU to that of standard AES, AES-NI and find that when the size of input data is different, the implementations with different parameters setting achieve the best performance. So we provide several advices about how to implement CBC-AES on GPU aiming at different size of input data. In particular, our best performance of experiments on GPU(NVIDIA Tesla K40m) is about 112 times faster than the implementation of AES on CPU (Intel Xeon E5-2650) by using our optimization method."
]
} |
1902.05183 | 2913499274 | In robotic surgery, pattern cutting through a deformable material is a challenging research field. The cutting procedure requires a robot to concurrently manipulate a scissor and a gripper to cut through a predefined contour trajectory on the deformable sheet. The gripper ensures the cutting accuracy by nailing a point on the sheet and continuously tensioning the pinch point to different directions while the scissor is in action. The goal is to find a pinch point and a corresponding tensioning policy to minimize damage to the material and increase cutting accuracy measured by the symmetric difference between the predefined contour and the cut contour. Previous study considers finding one fixed pinch point during the course of cutting, which is inaccurate and unsafe when the contour trajectory is complex. In this paper, we examine the soft tissue cutting task by using multiple pinch points, which imitates human operations while cutting. This approach, however, does not require the use of a multi-gripper robot. We use a deep reinforcement learning algorithm to find an optimal tensioning policy of a pinch point. Simulation results show that the multi-point approach outperforms the state-of-the-art method in soft pattern cutting task with respect to both accuracy and reliability. | Reinforcement learning (RL) has become a promising approach to modeling an autonomous agent @cite_37 @cite_17 @cite_32 @cite_23 @cite_34 . RL has the abilities to mimic human learning behaviors to maximize the long-term reward. As a result, RL enables a robot to learn on its own and partially eliminates the existence of experts. Examples of these agents are the box-pushing robot @cite_7 , pole-balancing humanoid @cite_33 , helicopter control @cite_29 , soccer-playing agent @cite_35 , and table tennis playing agent @cite_3 . Furthermore, RL has been utilized in a significant number of research interests in surgical tasks @cite_26 @cite_36 . For example, Chen @cite_2 propose a data-driven workflow that combines @cite_4 with RL. The workflow encodes the inverse kinematics using trajectories from human demonstrations. Finally, RL is used to minimize the noise and adapt the learned inverse kinematics to the online environment. | {
"cite_N": [
"@cite_35",
"@cite_37",
"@cite_26",
"@cite_4",
"@cite_33",
"@cite_7",
"@cite_36",
"@cite_29",
"@cite_32",
"@cite_3",
"@cite_23",
"@cite_2",
"@cite_34",
"@cite_17"
],
"mid": [
"2127412976",
"2908261578",
"2605447266",
"1684361744",
"2148112459",
"2018346149",
"",
"1949804828",
"",
"2012204020",
"",
"2313922182",
"",
""
],
"abstract": [
"Batch reinforcement learning methods provide a powerful framework for learning efficiently and effectively in autonomous robots. The paper reviews some recent work of the authors aiming at the successful application of reinforcement learning in a challenging and complex domain. It discusses several variants of the general batch learning framework, particularly tailored to the use of multilayer perceptrons to approximate value functions over continuous state spaces. The batch learning framework is successfully used to learn crucial skills in our soccer-playing robots participating in the RoboCup competitions. This is demonstrated on three different case studies.",
"Reinforcement learning (RL) algorithms have been around for decades and employed to solve various sequential decision-making problems. These algorithms however have faced great challenges when dealing with high-dimensional environments. The recent development of deep learning has enabled RL methods to drive optimal policies for sophisticated and capable agents, which can perform efficiently in these challenging environments. This paper addresses an important aspect of deep RL related to situations that require multiple agents to communicate and cooperate to solve complex tasks. A survey of different approaches to problems related to multi-agent deep RL (MADRL) is presented, including non-stationarity, partial observability, continuous state and action spaces, multi-agent training schemes, multi-agent transfer learning. The merits and demerits of the reviewed methods will be analyzed and discussed, with their corresponding applications explored. It is envisaged that this review provides insights about various MADRL methods and can lead to future development of more robust and highly useful multi-agent learning methods for solving real-world problems.",
"In order to get natural and intuitive physical interaction in the pose adjustment of the minimally invasive surgery manipulator, a hybrid variable admittance model based on Fuzzy Sarsa(λ)-learning is proposed in this paper. The proposed model provides continuous variable virtual damping to the admittance controller to respond to human intentions, and it effectively enhances the comfort level during the task execution by modifying the generated virtual damping dynamically. A fuzzy partition defined over the state space is used to capture the characteristics of the operator in physical human-robot interaction. For the purpose of maximizing the performance index in the long run, according to the identification of the current state input, the virtual damping compensations are determined by a trained strategy which can be learned through the experience generated from interaction with humans, and the influence caused by humans and the changing dynamics in the robot are also considered in the learning process. To evaluate the performance of the proposed model, some comparative experiments in joint space are conducted on our experimental minimally invasive surgical manipulator.",
"Also referred to as learning by imitation, tutelage, or apprenticeship learning, Programming by Demonstration (PbD) develops methods by which new skills can be transmitted to a robot. This book examines methods by which robots learn new skills through human guidance. Taking a practical perspective, it covers a broad range of applications, including service robots. The text addresses the challenges involved in investigating methods by which PbD is used to provide robots with a generic and adaptive model of control. Drawing on findings from robot control, human-robot interaction, applied machine learning, artificial intelligence, and developmental and cognitive psychology, the book contains a large set of didactic and illustrative examples. Practical and comprehensive machine learning source codes are available on the books companion website: http: www.programming-by-demonstration.org",
"By now it is widely accepted that learning a task from scratch, i.e., without any prior knowledge, is a daunting undertaking. Humans, however, rarely attempt to learn from scratch. They extract initial biases as well as strategies how to approach a learning problem from instructions and or demonstrations of other humans. For teaming control, this paper investigates how learning from demonstration can be applied in the context of reinforcement learning. We consider priming the Q-function, the value function, the policy, and the model of the task dynamics as possible areas where demonstrations can speed up learning. In general nonlinear learning problems, only model-based reinforcement learning shows significant speed-up after a demonstration, while in the special case of linear quadratic regulator (LQR) problems, all methods profit from the demonstration. In an implementation of pole balancing on a complex anthropomorphic robot arm, we demonstrate that, when facing the complexities of real signal processing, model-based reinforcement learning offers the most robustness for LQR problems. Using the suggested methods, the robot learns pole balancing in just a single trial after a 30 second long demonstration of the human instructor.",
"In the future, robotic surgical assistants may assist surgeons by performing specific subtasks such as retraction and suturing to reduce surgeon tedium and reduce the duration of some operations. We propose an apprenticeship learning approach that has potential to allow robotic surgical assistants to autonomously execute specific trajectories with superhuman performance in terms of speed and smoothness. In the first step, we record a set of trajectories using human-guided backdriven motions of the robot. These are then analyzed to extract a smooth reference trajectory, which we execute at gradually increasing speeds using a variant of iterative learning control. We evaluate this approach on two representative tasks using the Berkeley Surgical Robots: a figure eight trajectory and a two handed knot-tie, a tedious suturing sub-task required in many surgical procedures. Results suggest that the approach enables (i) rapid learning of trajectories, (ii) smoother trajectories than the human-guided trajectories, and (iii) trajectories that are 7 to 10 times faster than the best human-guided trajectories.",
"",
"Helicopters have highly stochastic, nonlinear, dynamics, and autonomous helicopter flight is widely regarded to be a challenging control problem. As helicopters are highly unstable at low speeds, it is particularly difficult to design controllers for low speed aerobatic maneuvers. In this paper, we describe a successful application of reinforcement learning to designing a controller for sustained inverted flight on an autonomous helicopter. Using data collected from the helicopter in flight, we began by learning a stochastic, nonlinear model of the helicopter’s dynamics. Then, a reinforcement learning algorithm was applied to automatically learn a controller for autonomous inverted hovering. Finally, the resulting controller was successfully tested on our autonomous helicopter platform.",
"",
"Learning new motor tasks from physical interactions is an important goal for both robotics and machine learning. However, when moving beyond basic skills, most monolithic machine learning approaches fail to scale. For more complex skills, methods that are tailored for the domain of skill learning are needed. In this paper, we take the task of learning table tennis as an example and present a new framework that allows a robot to learn cooperative table tennis from physical interaction with a human. The robot first learns a set of elementary table tennis hitting movements from a human table tennis teacher by kinesthetic teach-in, which is compiled into a set of motor primitives represented by dynamical systems. The robot subsequently generalizes these movements to a wider range of situations using our mixture of motor primitives approach. The resulting policy enables the robot to select appropriate motor primitives as well as to generalize between them. Finally, the robot plays with a human table tennis partner and learns online to improve its behavior. We show that the resulting setup is capable of playing table tennis using an anthropomorphic robot arm.",
"",
"Flexible manipulators such as tendon-driven serpentine manipulators perform better than traditional rigid ones in minimally invasive surgical tasks, including navigation in confined space through key-hole like incisions. However, due to the inherent nonlinearities and model uncertainties, motion control of such manipulators becomes extremely challenging. In this work, a hybrid framework combining Programming by Demonstration (PbD) and reinforcement learning is proposed to solve this problem. Gaussian Mixture Models (GMM), Gaussian Mixture Regression (GMR) and linear regression are used to learn the inverse kinematic model of the manipulator from human demonstrations. The learned model is used as nominal model to calculate the output end-effector trajectories of the manipulator. Two surgical tasks are performed to demonstrate the effectiveness of reinforcement learning: tube insertion and circle following. Gaussian noise is introduced to the standard model and the disturbed models are fed to the manipulator to calculate the actuator input with respect to the task specific end-effector trajectories. An expectation maximization (E-M) based reinforcement learning algorithm is used to update the disturbed model with returns from rollouts. Simulation results have verified that the disturbed model can be converged to the standard one and the tracking accuracy is enhanced.",
"",
""
]
} |
1902.05183 | 2913499274 | In robotic surgery, pattern cutting through a deformable material is a challenging research field. The cutting procedure requires a robot to concurrently manipulate a scissor and a gripper to cut through a predefined contour trajectory on the deformable sheet. The gripper ensures the cutting accuracy by nailing a point on the sheet and continuously tensioning the pinch point to different directions while the scissor is in action. The goal is to find a pinch point and a corresponding tensioning policy to minimize damage to the material and increase cutting accuracy measured by the symmetric difference between the predefined contour and the cut contour. Previous study considers finding one fixed pinch point during the course of cutting, which is inaccurate and unsafe when the contour trajectory is complex. In this paper, we examine the soft tissue cutting task by using multiple pinch points, which imitates human operations while cutting. This approach, however, does not require the use of a multi-gripper robot. We use a deep reinforcement learning algorithm to find an optimal tensioning policy of a pinch point. Simulation results show that the multi-point approach outperforms the state-of-the-art method in soft pattern cutting task with respect to both accuracy and reliability. | Recent breakthrough @cite_13 combines neural networks with RL (deep RL), which enables traditional RL methods to work properly in high-dimensional environments. Typically, Thananjeyan @cite_1 use a deep RL algorithm (TRPO) to learn the tensioning policy from a pinch point in pattern cutting task. The tensioning problem is described as a @cite_24 where each action moves the gripper 1mm along one of four directions in the 2D space. A state of the environment is a state of the deformable sheet, which is represented by a rectangular mesh of point masses. The goal is to minimize the symmetric difference between the ideal contour and the achieved contour cut. This approach, however, uses a single pinch point to complete the pattern cutting task. In this paper, we examine the use of multiple pinch points over the course of cutting. Finally, we compare the cutting accuracy of our proposed scheme with two baseline methods described in @cite_1 : 1) the non-tensioned scheme and 2) SDRLT. Section and Section present the proposed scheme in more details. | {
"cite_N": [
"@cite_1",
"@cite_13",
"@cite_24"
],
"mid": [
"2736750746",
"2145339207",
"2121863487"
],
"abstract": [
"In the Fundamentals of Laparoscopic Surgery (FLS) standard medical training regimen, the Pattern Cutting task requires residents to demonstrate proficiency by maneuvering two tools, surgical scissors and tissue gripper, to accurately cut a circular pattern on surgical gauze suspended at the corners. Accuracy of cutting depends on tensioning, wherein the gripper pinches a point on the gauze in R3 and pulls to induce and maintain tension in the material as cutting proceeds. An automated tensioning policy maps the current state of the gauze to output a direction of pulling as an action. The optimal tensioning policy depends on both the choice of pinch point and cutting trajectory. We explore the problem of learning a tensioning policy conditioned on specific cutting trajectories. Every timestep, we allow the gripper to react to the deformation of the gauze and progress of the cutting trajectory with a translation unit vector along an allowable set of directions. As deformation is difficult to analytically model and explicitly observe, we leverage deep reinforcement learning with direct policy search methods to learn tensioning policies using a finite-element simulator and then transfer them to a physical system. We compare the Deep RL tensioning policies with fixed and analytic (opposing the error vector with a fixed pinch point) policies on a set of 17 open and closed curved contours in simulation and 4 patterns in physical experiments with the da Vinci Research Kit (dVRK). Our simulation results suggest that learning to tension with Deep RL can significantly improve performance and robustness to noise and external forces.",
"An artificial agent is developed that learns to play a diverse range of classic Atari 2600 computer games directly from sensory experience, achieving a performance comparable to that of an expert human player; this work paves the way to building general-purpose learning algorithms that bridge the divide between perception and action.",
"Reinforcement learning, one of the most active research areas in artificial intelligence, is a computational approach to learning whereby an agent tries to maximize the total amount of reward it receives when interacting with a complex, uncertain environment. In Reinforcement Learning, Richard Sutton and Andrew Barto provide a clear and simple account of the key ideas and algorithms of reinforcement learning. Their discussion ranges from the history of the field's intellectual foundations to the most recent developments and applications. The only necessary mathematical background is familiarity with elementary concepts of probability. The book is divided into three parts. Part I defines the reinforcement learning problem in terms of Markov decision processes. Part II provides basic solution methods: dynamic programming, Monte Carlo methods, and temporal-difference learning. Part III presents a unified view of the solution methods and incorporates artificial neural networks, eligibility traces, and planning; the two final chapters present case studies and consider the future of reinforcement learning."
]
} |
1902.04981 | 2914095169 | Abstract A promising direction in deep learning research consists in learning representations and simultaneously discovering cluster structure in unlabeled data by optimizing a discriminative loss function. As opposed to supervised deep learning, this line of research is in its infancy, and how to design and optimize suitable loss functions to train deep neural networks for clustering is still an open question. Our contribution to this emerging field is a new deep clustering network that leverages the discriminative power of information-theoretic divergence measures, which have been shown to be effective in traditional clustering. We propose a novel loss function that incorporates geometric regularization constraints, thus avoiding degenerate structures of the resulting clustering partition. Experiments on synthetic benchmarks and real datasets show that the proposed network achieves competitive performance with respect to other state-of-the-art methods, scales well to large datasets, and does not require pre-training steps. | Common approaches to unsupervised deep learning include methods based on deep belief networks, autoencoders, and generative adversarial networks @cite_51 @cite_23 . These methods have been mainly used for unsupervised pre-training @cite_10 . Deep belief networks were the first of these models to be proposed and consist of stacked restricted Boltzmann machines that are trained in a greedy fashion @cite_61 . Once trained, deep belief networks can be used to initialize neural networks. | {
"cite_N": [
"@cite_61",
"@cite_10",
"@cite_51",
"@cite_23"
],
"mid": [
"2136922672",
"2138857742",
"2163922914",
"2099471712"
],
"abstract": [
"We show how to use \"complementary priors\" to eliminate the explaining-away effects that make inference difficult in densely connected belief nets that have many hidden layers. Using complementary priors, we derive a fast, greedy algorithm that can learn deep, directed belief networks one layer at a time, provided the top two layers form an undirected associative memory. The fast, greedy algorithm is used to initialize a slower learning procedure that fine-tunes the weights using a contrastive version of the wake-sleep algorithm. After fine-tuning, a network with three hidden layers forms a very good generative model of the joint distribution of handwritten digit images and their labels. This generative model gives better digit classification than the best discriminative learning algorithms. The low-dimensional manifolds on which the digits lie are modeled by long ravines in the free-energy landscape of the top-level associative memory, and it is easy to explore these ravines by using the directed connections to display what the associative memory has in mind.",
"Much recent research has been devoted to learning algorithms for deep architectures such as Deep Belief Networks and stacks of auto-encoder variants, with impressive results obtained in several areas, mostly on vision and language data sets. The best results obtained on supervised learning tasks involve an unsupervised learning component, usually in an unsupervised pre-training phase. Even though these new algorithms have enabled training deep models, many questions remain as to the nature of this difficult learning problem. The main question investigated here is the following: how does unsupervised pre-training work? Answering this questions is important if learning in deep architectures is to be further improved. We propose several explanatory hypotheses and test them through extensive simulations. We empirically show the influence of pre-training with respect to architecture depth, model capacity, and number of training examples. The experiments confirm and clarify the advantage of unsupervised pre-training. The results suggest that unsupervised pre-training guides the learning towards basins of attraction of minima that support better generalization from the training data set; the evidence from these results supports a regularization explanation for the effect of pre-training.",
"The success of machine learning algorithms generally depends on data representation, and we hypothesize that this is because different representations can entangle and hide more or less the different explanatory factors of variation behind the data. Although specific domain knowledge can be used to help design representations, learning with generic priors can also be used, and the quest for AI is motivating the design of more powerful representation-learning algorithms implementing such priors. This paper reviews recent work in the area of unsupervised feature learning and deep learning, covering advances in probabilistic models, autoencoders, manifold learning, and deep networks. This motivates longer term unanswered questions about the appropriate objectives for learning good representations, for computing representations (i.e., inference), and the geometrical connections between representation learning, density estimation, and manifold learning.",
"We propose a new framework for estimating generative models via an adversarial process, in which we simultaneously train two models: a generative model G that captures the data distribution, and a discriminative model D that estimates the probability that a sample came from the training data rather than G. The training procedure for G is to maximize the probability of D making a mistake. This framework corresponds to a minimax two-player game. In the space of arbitrary functions G and D, a unique solution exists, with G recovering the training data distribution and D equal to ½ everywhere. In the case where G and D are defined by multilayer perceptrons, the entire system can be trained with backpropagation. There is no need for any Markov chains or unrolled approximate inference networks during either training or generation of samples. Experiments demonstrate the potential of the framework through qualitative and quantitative evaluation of the generated samples."
]
} |
1902.04981 | 2914095169 | Abstract A promising direction in deep learning research consists in learning representations and simultaneously discovering cluster structure in unlabeled data by optimizing a discriminative loss function. As opposed to supervised deep learning, this line of research is in its infancy, and how to design and optimize suitable loss functions to train deep neural networks for clustering is still an open question. Our contribution to this emerging field is a new deep clustering network that leverages the discriminative power of information-theoretic divergence measures, which have been shown to be effective in traditional clustering. We propose a novel loss function that incorporates geometric regularization constraints, thus avoiding degenerate structures of the resulting clustering partition. Experiments on synthetic benchmarks and real datasets show that the proposed network achieves competitive performance with respect to other state-of-the-art methods, scales well to large datasets, and does not require pre-training steps. | Although several types of autoencoders have been proposed, all share a common underlying architecture consisting of an encoding and a decoding layer. The encoder is responsible for producing a hidden representation; the decoder re-generates inputs from the hidden representation. Both can efficiently be learned using backpropagation, by minimizing the reconstruction loss between original input and decoder output. Variations include, among others, denoising autoencoders @cite_20 , which regularize the original autoencoder model by adding noise to inputs and then changing the objective to both include reconstruction and denoising, contractive autoencoders @cite_21 , and more recently autoencoders that are regularized by preserving similarities in input space @cite_39 . Variational autoencoders @cite_50 have been used recently for several unsupervised tasks, such as image generation @cite_34 and segmentation @cite_24 . This approach assumes that data are generated from directed graphical models and uses a variational approach to learn latent representations. | {
"cite_N": [
"@cite_21",
"@cite_39",
"@cite_24",
"@cite_50",
"@cite_34",
"@cite_20"
],
"mid": [
"2218318129",
"2586314446",
"2188365844",
"",
"1850742715",
"2025768430"
],
"abstract": [
"We present in this paper a novel approach for training deterministic auto-encoders. We show that by adding a well chosen penalty term to the classical reconstruction cost function, we can achieve results that equal or surpass those attained by other regularized auto-encoders as well as denoising auto-encoders on a range of datasets. This penalty term corresponds to the Frobenius norm of the Jacobian matrix of the encoder activations with respect to the input. We show that this penalty term results in a localized space contraction which in turn yields robust features on the activation layer. Furthermore, we show how this penalty term is related to both regularized auto-encoders and denoising auto-encoders and how it can be seen as a link between deterministic and non-deterministic auto-encoders. We find empirically that this penalty helps to carve a representation that better captures the local directions of variation dictated by the data, corresponding to a lower-dimensional non-linear manifold, while being more invariant to the vast majority of directions orthogonal to the manifold. Finally, we show that by using the learned features to initialize a MLP, we achieve state of the art classification error on a range of datasets, surpassing other methods of pretraining.",
"In this paper we introduce the deep kernelized autoencoder, a neural network model that allows an explicit approximation of (i) the mapping from an input space to an arbitrary, user-specified kernel space and (ii) the back-projection from such a kernel space to input space. The proposed method is based on traditional autoencoders and is trained through a new unsupervised loss function. During training, we optimize both the reconstruction accuracy of input samples and the alignment between a kernel matrix given as prior and the inner products of the hidden representations computed by the autoencoder. Kernel alignment provides control over the hidden representation learned by the autoencoder. Experiments have been performed to evaluate both reconstruction and kernel alignment performance. Additionally, we applied our method to emulate kPCA on a denoising task obtaining promising results.",
"Supervised deep learning has been successfully applied to many recognition problems. Although it can approximate a complex many-to-one function well when a large amount of training data is provided, it is still challenging to model complex structured output representations that effectively perform probabilistic inference and make diverse predictions. In this work, we develop a deep conditional generative model for structured output prediction using Gaussian latent variables. The model is trained efficiently in the framework of stochastic gradient variational Bayes, and allows for fast prediction using stochastic feed-forward inference. In addition, we provide novel strategies to build robust structured prediction algorithms, such as input noise-injection and multi-scale prediction objective at training. In experiments, we demonstrate the effectiveness of our proposed algorithm in comparison to the deterministic deep neural network counterparts in generating diverse but realistic structured output predictions using stochastic inference. Furthermore, the proposed training methods are complimentary, which leads to strong pixel-level object segmentation and semantic labeling performance on Caltech-UCSD Birds 200 and the subset of Labeled Faces in the Wild dataset.",
"",
"This paper introduces the Deep Recurrent Attentive Writer (DRAW) neural network architecture for image generation. DRAW networks combine a novel spatial attention mechanism that mimics the foveation of the human eye, with a sequential variational auto-encoding framework that allows for the iterative construction of complex images. The system substantially improves on the state of the art for generative models on MNIST, and, when trained on the Street View House Numbers dataset, it generates images that cannot be distinguished from real data with the naked eye.",
"Previous work has shown that the difficulties in learning deep generative or discriminative models can be overcome by an initial unsupervised learning step that maps inputs to useful intermediate representations. We introduce and motivate a new training principle for unsupervised learning of a representation based on the idea of making the learned representations robust to partial corruption of the input pattern. This approach can be used to train autoencoders, and these denoising autoencoders can be stacked to initialize deep architectures. The algorithm can be motivated from a manifold learning and information theoretic perspective or from a generative model perspective. Comparative experiments clearly show the surprising advantage of corrupting the input of autoencoders on a pattern classification benchmark suite."
]
} |
1902.04981 | 2914095169 | Abstract A promising direction in deep learning research consists in learning representations and simultaneously discovering cluster structure in unlabeled data by optimizing a discriminative loss function. As opposed to supervised deep learning, this line of research is in its infancy, and how to design and optimize suitable loss functions to train deep neural networks for clustering is still an open question. Our contribution to this emerging field is a new deep clustering network that leverages the discriminative power of information-theoretic divergence measures, which have been shown to be effective in traditional clustering. We propose a novel loss function that incorporates geometric regularization constraints, thus avoiding degenerate structures of the resulting clustering partition. Experiments on synthetic benchmarks and real datasets show that the proposed network achieves competitive performance with respect to other state-of-the-art methods, scales well to large datasets, and does not require pre-training steps. | Adversarial generative models @cite_23 are more recent approaches to unsupervised deep learning. Here, two networks are trained: one is responsible for discriminating between real and generated images; the other is responsible for generating realistic-enough images to confuse the first network. | {
"cite_N": [
"@cite_23"
],
"mid": [
"2099471712"
],
"abstract": [
"We propose a new framework for estimating generative models via an adversarial process, in which we simultaneously train two models: a generative model G that captures the data distribution, and a discriminative model D that estimates the probability that a sample came from the training data rather than G. The training procedure for G is to maximize the probability of D making a mistake. This framework corresponds to a minimax two-player game. In the space of arbitrary functions G and D, a unique solution exists, with G recovering the training data distribution and D equal to ½ everywhere. In the case where G and D are defined by multilayer perceptrons, the entire system can be trained with backpropagation. There is no need for any Markov chains or unrolled approximate inference networks during either training or generation of samples. Experiments demonstrate the potential of the framework through qualitative and quantitative evaluation of the generated samples."
]
} |
1902.04981 | 2914095169 | Abstract A promising direction in deep learning research consists in learning representations and simultaneously discovering cluster structure in unlabeled data by optimizing a discriminative loss function. As opposed to supervised deep learning, this line of research is in its infancy, and how to design and optimize suitable loss functions to train deep neural networks for clustering is still an open question. Our contribution to this emerging field is a new deep clustering network that leverages the discriminative power of information-theoretic divergence measures, which have been shown to be effective in traditional clustering. We propose a novel loss function that incorporates geometric regularization constraints, thus avoiding degenerate structures of the resulting clustering partition. Experiments on synthetic benchmarks and real datasets show that the proposed network achieves competitive performance with respect to other state-of-the-art methods, scales well to large datasets, and does not require pre-training steps. | Clustering is a classic information processing problem, particularly important in machine learning @cite_7 @cite_66 @cite_11 @cite_26 @cite_42 . Countless approaches exist for clustering, with mean shift @cite_54 , @math -means and expectation--maximization algorithms @cite_65 , being some of the most well-known ones. In the last decade, played a prominent role in the field, see for instance @cite_0 @cite_58 @cite_22 @cite_25 @cite_4 . Spectral clustering exploits the spectrum of similarity matrices to partition input data. Although these methods have demonstrated good performance in complex problems, they suffer from lack of scalability with respect to the number of input data points; cubic computational complexity for eigensolvers and quadratic complexity in terms of memory occupation. Attempts to solve these problems have been made by designing approximations or employing different optimization techniques @cite_3 @cite_49 @cite_6 . | {
"cite_N": [
"@cite_26",
"@cite_25",
"@cite_22",
"@cite_7",
"@cite_4",
"@cite_54",
"@cite_42",
"@cite_65",
"@cite_3",
"@cite_6",
"@cite_0",
"@cite_49",
"@cite_58",
"@cite_66",
"@cite_11"
],
"mid": [
"2809506935",
"2132914434",
"2131828344",
"2011430131",
"2129793592",
"2067191022",
"2772069281",
"1531910981",
"1986007546",
"2134370969",
"2149803014",
"2511280622",
"2165874743",
"2165835468",
"2024096326"
],
"abstract": [
"In this paper, we make a multiview extension of the spectral rotation technique raised in single view spectral clustering research. Since spectral rotation is closely related to the Procrustes Analysis for points matching, we point out that classical Procrustes Average approach can be used for multiview clustering. Besides, we show that direct applying Procrustes Average (PA) in multiview tasks may not be optimal theoretically and empirically, since it does not take the clustering capacity differences of different views into consideration. Other than that, we propose an Adaptively Weighted Procrustes (AWP) approach to overcome the aforementioned deficiency. Our new AWP weights views with their clustering capacities and forms a weighted Procrustes Average problem accordingly. The optimization algorithm to solve the new model is computational complexity analyzed and convergence guaranteed. Experiments on five real-world datasets demonstrate the effectiveness and efficiency of the new models.",
"In recent years, spectral clustering has become one of the most popular modern clustering algorithms. It is simple to implement, can be solved efficiently by standard linear algebra software, and very often outperforms traditional clustering algorithms such as the k-means algorithm. On the first glance spectral clustering appears slightly mysterious, and it is not obvious to see why it works at all and what it really does. The goal of this tutorial is to give some intuition on those questions. We describe different graph Laplacians and their basic properties, present the most common spectral clustering algorithms, and derive those algorithms from scratch by several different approaches. Advantages and disadvantages of the different spectral clustering algorithms are discussed.",
"Spectral clustering (SC) methods have been successfully applied to many real-world applications. The success of these SC methods is largely based on the manifold assumption, namely, that two nearby data points in the high-density region of a low-dimensional data manifold have the same cluster label. However, such an assumption might not always hold on high-dimensional data. When the data do not exhibit a clear low-dimensional manifold structure (e.g., high-dimensional and sparse data), the clustering performance of SC will be degraded and become even worse than K -means clustering. In this paper, motivated by the observation that the true cluster assignment matrix for high-dimensional data can be always embedded in a linear space spanned by the data, we propose the spectral embedded clustering (SEC) framework, in which a linearity regularization is explicitly added into the objective function of SC methods. More importantly, the proposed SEC framework can naturally deal with out-of-sample data. We also present a new Laplacian matrix constructed from a local regression of each pattern and incorporate it into our SEC framework to capture both local and global discriminative information for clustering. Comprehensive experiments on eight real-world high-dimensional datasets demonstrate the effectiveness and advantages of our SEC framework over existing SC methods and K-means-based clustering methods. Our SEC framework significantly outperforms SC using the Nystrom algorithm on unseen data.",
"Organizing data into sensible groupings is one of the most fundamental modes of understanding and learning. As an example, a common scheme of scientific classification puts organisms into a system of ranked taxa: domain, kingdom, phylum, class, etc. Cluster analysis is the formal study of methods and algorithms for grouping, or clustering, objects according to measured or perceived intrinsic characteristics or similarity. Cluster analysis does not use category labels that tag objects with prior identifiers, i.e., class labels. The absence of category information distinguishes data clustering (unsupervised learning) from classification or discriminant analysis (supervised learning). The aim of clustering is to find structure in data and is therefore exploratory in nature. Clustering has a long and rich history in a variety of scientific fields. One of the most popular and simple clustering algorithms, K-means, was first published in 1955. In spite of the fact that K-means was proposed over 50 years ago and thousands of clustering algorithms have been published since then, K-means is still widely used. This speaks to the difficulty in designing a general purpose clustering algorithm and the ill-posed problem of clustering. We provide a brief overview of clustering, summarize well known clustering methods, discuss the major challenges and key issues in designing clustering algorithms, and point out some of the emerging and useful research directions, including semi-supervised clustering, ensemble clustering, simultaneous feature selection during data clustering, and large scale data clustering.",
"In this paper, we propose a new image clustering algorithm, referred to as clustering using local discriminant models and global integration (LDMGI). To deal with the data points sampled from a nonlinear manifold, for each data point, we construct a local clique comprising this data point and its neighboring data points. Inspired by the Fisher criterion, we use a local discriminant model for each local clique to evaluate the clustering performance of samples within the local clique. To obtain the clustering result, we further propose a unified objective function to globally integrate the local models of all the local cliques. With the unified objective function, spectral relaxation and spectral rotation are used to obtain the binary cluster indicator matrix for all the samples. We show that LDMGI shares a similar objective function with the spectral clustering (SC) algorithms, e.g., normalized cut (NCut). In contrast to NCut in which the Laplacian matrix is directly calculated based upon a Gaussian function, a new Laplacian matrix is learnt in LDMGI by exploiting both manifold structure and local discriminant information. We also prove that K-means and discriminative K-means (DisKmeans) are both special cases of LDMGI. Extensive experiments on several benchmark image datasets demonstrate the effectiveness of LDMGI. We observe in the experiments that LDMGI is more robust to algorithmic parameter, when compared with NCut. Thus, LDMGI is more appealing for the real image clustering applications in which the ground truth is generally not available for tuning algorithmic parameters.",
"A general non-parametric technique is proposed for the analysis of a complex multimodal feature space and to delineate arbitrarily shaped clusters in it. The basic computational module of the technique is an old pattern recognition procedure: the mean shift. For discrete data, we prove the convergence of a recursive mean shift procedure to the nearest stationary point of the underlying density function and, thus, its utility in detecting the modes of the density. The relation of the mean shift procedure to the Nadaraya-Watson estimator from kernel regression and the robust M-estimators; of location is also established. Algorithms for two low-level vision tasks discontinuity-preserving smoothing and image segmentation - are described as applications. In these algorithms, the only user-set parameter is the resolution of the analysis, and either gray-level or color images are accepted as input. Extensive experimental results illustrate their excellent performance.",
"A new clustering ensemble based on kNN mode seeking is proposed.The algorithm is robust with respect to hyperparametersno manual tuning needed.The algorithm is faster than the state-of-the art and able to handle high-dimensional data sets. In this paper we present a new algorithm for parameter-free clustering by mode seeking. Mode seeking, especially in the form of the mean shift algorithm, is a widely used strategy for clustering data, but at the same time prone to poor performance if the parameters are not chosen correctly. We propose to form a clustering ensemble consisting of repeated and bootstrapped runs of the recent kNN mode seeking algorithm, an algorithm which is faster than ordinary mean shift and more suited for high dimensional data. This creates a robust mode seeking clustering algorithm with respect to the choice of parameters and high dimensional input spaces, while at the same inheriting all other strengths of mode seeking in general. We demonstrate promising results on a number of synthetic and real data sets.",
"Research on the problem of clustering tends to be fragmented across the pattern recognition, database, data mining, and machine learning communities. Addressing this problem in a unified way, Data Clustering: Algorithms and Applications provides complete coverage of the entire area of clustering, from basic methods to more refined and complex data clustering approaches. It pays special attention to recent issues in graphs, social networks, and other domains. The book focuses on three primary aspects of data clustering: Methods, describing key techniques commonly used for clustering, such as feature selection, agglomerative clustering, partitional clustering, density-based clustering, probabilistic clustering, grid-based clustering, spectral clustering, and nonnegative matrix factorization Domains, covering methods used for different domains of data, such as categorical data, text data, multimedia data, graph data, biological data, stream data, uncertain data, time series clustering, high-dimensional clustering, and big data Variations and Insights, discussing important variations of the clustering process, such as semisupervised clustering, interactive clustering, multiview clustering, cluster ensembles, and cluster validation In this book, top researchers from around the world explore the characteristics of clustering problems in a variety of application areas. They also explain how to glean detailed insight from the clustering processincluding how to verify the quality of the underlying clustersthrough supervision, human intervention, or the automated generation of alternative clusters.",
"Kernel k-means and spectral clustering have both been used to identify clusters that are non-linearly separable in input space. Despite significant research, these methods have remained only loosely related. In this paper, we give an explicit theoretical connection between them. We show the generality of the weighted kernel k-means objective function, and derive the spectral clustering objective of normalized cut as a special case. Given a positive definite similarity matrix, our results lead to a novel weighted kernel k-means algorithm that monotonically decreases the normalized cut. This has important implications: a) eigenvector-based algorithms, which can be computationally prohibitive, are not essential for minimizing normalized cuts, b) various techniques, such as local search and acceleration schemes, may be used to improve the quality as well as speed of kernel k-means. Finally, we present results on several interesting data sets, including diametrical clustering of large gene-expression matrices and a handwriting recognition data set.",
"Spectral clustering refers to a flexible class of clustering procedures that can produce high-quality clusterings on small data sets but which has limited applicability to large-scale problems due to its computational complexity of O(n3) in general, with n the number of data points. We extend the range of spectral clustering by developing a general framework for fast approximate spectral clustering in which a distortion-minimizing local transformation is first applied to the data. This framework is based on a theoretical analysis that provides a statistical characterization of the effect of local distortion on the mis-clustering rate. We develop two concrete instances of our general framework, one based on local k-means clustering (KASP) and one based on random projection trees (RASP). Extensive experiments show that these algorithms can achieve significant speedups with little degradation in clustering accuracy. Specifically, our algorithms outperform k-means by a large margin in terms of accuracy, and run several times faster than approximate spectral clustering based on the Nystrom method, with comparable accuracy and significantly smaller memory footprint. Remarkably, our algorithms make it possible for a single machine to spectral cluster data sets with a million observations within several minutes.",
"We introduce kernel entropy component analysis (kernel ECA) as a new method for data transformation and dimensionality reduction. Kernel ECA reveals structure relating to the Renyi entropy of the input space data set, estimated via a kernel matrix using Parzen windowing. This is achieved by projections onto a subset of entropy preserving kernel principal component analysis (kernel PCA) axes. This subset does not need, in general, to correspond to the top eigenvalues of the kernel matrix, in contrast to the dimensionality reduction using kernel PCA. We show that kernel ECA may produce strikingly different transformed data sets compared to kernel PCA, with a distinct angle-based structure. A new spectral clustering algorithm utilizing this structure is developed with positive results. Furthermore, kernel ECA is shown to be an useful alternative for pattern denoising.",
"The cost of computing the spectrum of Laplacian matrices hinders the application of spectral clustering to large data sets. While approximations recover computational tractability, they can potentially affect clustering performance. This paper proposes a practical approach to learn spectral clustering based on adaptive stochastic gradient optimization. Crucially, the proposed approach recovers the exact spectrum of Laplacian matrices in the limit of the iterations, and the cost of each iteration is linear in the number of samples. Extensive experimental validation on data sets with up to half a million samples demonstrate its scalability and its ability to outperform state-of-the-art approximate methods to learn spectral clustering for a given computational budget.",
"Despite many empirical successes of spectral clustering methods— algorithms that cluster points using eigenvectors of matrices derived from the data—there are several unresolved issues. First. there are a wide variety of algorithms that use the eigenvectors in slightly different ways. Second, many of these algorithms have no proof that they will actually compute a reasonable clustering. In this paper, we present a simple spectral clustering algorithm that can be implemented using a few lines of Matlab. Using tools from matrix perturbation theory, we analyze the algorithm, and give conditions under which it can be expected to do well. We also show surprisingly good experimental results on a number of challenging clustering problems.",
"Cluster analysis is aimed at classifying elements into categories on the basis of their similarity. Its applications range from astronomy to bioinformatics, bibliometrics, and pattern recognition. We propose an approach based on the idea that cluster centers are characterized by a higher density than their neighbors and by a relatively large distance from points with higher densities. This idea forms the basis of a clustering procedure in which the number of clusters arises intuitively, outliers are automatically spotted and excluded from the analysis, and clusters are recognized regardless of their shape and of the dimensionality of the space in which they are embedded. We demonstrate the power of the algorithm on several test cases.",
"In this paper, we propose a density-based clusters' representatives selection algorithm that identifies the most central patterns from the dense regions in the dataset. The method, which has been implemented using two different strategies, is applicable to input spaces with no trivial geometry. Our approach exploits a probability density function built through the Parzen estimator, which relies on a (not necessarily metric) dissimilarity measure. Being a representatives extractor a general-purpose algorithm, our method is obviously applicable in different contexts. However, to test the proposed procedure, we specifically consider the problem of initializing the k-means algorithm. We face problems defined on standard real-valued vectors, labeled graphs, and finally sequences of real-valued vectors and sequences of characters. The obtained results demonstrate the effectiveness of the proposed representative selection method with respect to other state-of-the-art solutions."
]
} |
1902.04981 | 2914095169 | Abstract A promising direction in deep learning research consists in learning representations and simultaneously discovering cluster structure in unlabeled data by optimizing a discriminative loss function. As opposed to supervised deep learning, this line of research is in its infancy, and how to design and optimize suitable loss functions to train deep neural networks for clustering is still an open question. Our contribution to this emerging field is a new deep clustering network that leverages the discriminative power of information-theoretic divergence measures, which have been shown to be effective in traditional clustering. We propose a novel loss function that incorporates geometric regularization constraints, thus avoiding degenerate structures of the resulting clustering partition. Experiments on synthetic benchmarks and real datasets show that the proposed network achieves competitive performance with respect to other state-of-the-art methods, scales well to large datasets, and does not require pre-training steps. | To the best of our knowledge, DEC @cite_14 is the method that is most closely related to our approach, as it is also founded on traditional clustering approaches. DEC simultaneously learns a feature representation as well as a cluster assignment in a two-step procedure. In the first step, soft assignments are computed between the data and cluster centroids based on a Student's t-distribution. Then, the parameters are optimized by matching soft assignments to a target distribution, which expresses confidence in assignments. The matching is performed by minimizing the Kullback-Leibler divergence. However, the effectiveness of DEC depends on a pre-training step implemented with autoencoders and does require explicit feature design to handle complex image data, e.g., Histogram of Oriented Gradients (HOG) features @cite_41 . | {
"cite_N": [
"@cite_41",
"@cite_14"
],
"mid": [
"2161969291",
"2964074409"
],
"abstract": [
"We study the question of feature sets for robust visual object recognition; adopting linear SVM based human detection as a test case. After reviewing existing edge and gradient based descriptors, we show experimentally that grids of histograms of oriented gradient (HOG) descriptors significantly outperform existing feature sets for human detection. We study the influence of each stage of the computation on performance, concluding that fine-scale gradients, fine orientation binning, relatively coarse spatial binning, and high-quality local contrast normalization in overlapping descriptor blocks are all important for good results. The new approach gives near-perfect separation on the original MIT pedestrian database, so we introduce a more challenging dataset containing over 1800 annotated human images with a large range of pose variations and backgrounds.",
"Clustering is central to many data-driven application domains and has been studied extensively in terms of distance functions and grouping algorithms. Relatively little work has focused on learning representations for clustering. In this paper, we propose Deep Embedded Clustering (DEC), a method that simultaneously learns feature representations and cluster assignments using deep neural networks. DEC learns a mapping from the data space to a lower-dimensional feature space in which it iteratively optimizes a clustering objective. Our experimental evaluations on image and text corpora show significant improvement over state-of-the-art methods."
]
} |
1902.05066 | 2921001691 | Multi-instance learning (MIL) deals with tasks where data consist of set of bags and each bag is represented by a set of instances. Only the bag labels are observed but the label for each instance is not available. Previous MIL studies typically assume that the training and test samples follow the same distribution, which is often violated in real-world applications. Existing methods address distribution changes by re-weighting the training data with the density ratio between the training and test samples. However, models are often trained without prior knowledge of the test distribution which renders existing methods inapplicable. Inspired by a connection between MIL and causal inference, we propose a novel framework for addressing distribution change in MIL without relying on the test distribution. Experimental results validate the effectiveness of our approach. | Multi-instance learning was first proposed for drug activity prediction @cite_10 . Since then, many algorithms have been proposed to solve the MIL problem, which can be roughly divided into two categories: one group of methods aim to directly solve the MIL problem in either instance level @cite_20 or bag level @cite_16 , and another category of methods transform MIL into single instance learning via bag embedding, among which MILES @cite_9 is a typical representative, and many methods @cite_1 @cite_14 have been proposed thereafter following this paradigm. | {
"cite_N": [
"@cite_14",
"@cite_9",
"@cite_1",
"@cite_16",
"@cite_10",
"@cite_20"
],
"mid": [
"2343974644",
"2098166271",
"2127608660",
"2133288557",
"2110119381",
"2108745803"
],
"abstract": [
"Multi-instance learning (MIL) has been widely applied to diverse applications involving complicated data objects, such as images and genes. However, most existing MIL algorithms can only handle small- or moderate-sized data. In order to deal with large-scale MIL problems, we propose MIL based on the vector of locally aggregated descriptors representation (miVLAD) and MIL based on the Fisher vector representation (miFV), two efficient and scalable MIL algorithms. They map the original MIL bags into new vector representations using their corresponding mapping functions. The new feature representations keep essential bag-level information, and at the same time lead to excellent MIL performances even when linear classifiers are used. Thanks to the low computational cost in the mapping step and the scalability of linear classifiers, miVLAD and miFV can handle large-scale MIL data efficiently and effectively. Experiments show that miVLAD and miFV not only achieve comparable accuracy rates with the state-of-the-art MIL algorithms, but also have hundreds of times faster speed. Moreover, we can regard the new miVLAD and miFV representations as multiview data, which improves the accuracy rates in most cases. In addition, our algorithms perform well even when they are used without parameter tuning (i.e., adopting the default parameters), which is convenient for practical MIL applications.",
"Multiple-instance problems arise from the situations where training class labels are attached to sets of samples (named bags), instead of individual samples within each bag (called instances). Most previous multiple-instance learning (MIL) algorithms are developed based on the assumption that a bag is positive if and only if at least one of its instances is positive. Although the assumption works well in a drug activity prediction problem, it is rather restrictive for other applications, especially those in the computer vision area. We propose a learning method, MILES (multiple-instance learning via embedded instance selection), which converts the multiple-instance learning problem to a standard supervised learning problem that does not impose the assumption relating instance labels to bag labels. MILES maps each bag into a feature space defined by the instances in the training bags via an instance similarity measure. This feature mapping often provides a large number of redundant or irrelevant features. Hence, 1-norm SVM is applied to select important features as well as construct classifiers simultaneously. We have performed extensive experiments. In comparison with other methods, MILES demonstrates competitive classification accuracy, high computation efficiency, and robustness to labeling uncertainty",
"Multiple instance learning (MIL) is a paradigm in supervised learning that deals with the classification of collections of instances called bags. Each bag contains a number of instances from which features are extracted. The complexity of MIL is largely dependent on the number of instances in the training data set. Since we are usually confronted with a large instance space even for moderately sized real-world data sets applications, it is important to design efficient instance selection techniques to speed up the training process without compromising the performance. In this paper, we address the issue of instance selection in MIL. We propose MILIS, a novel MIL algorithm based on adaptive instance selection. We do this in an alternating optimization framework by intertwining the steps of instance selection and classifier learning in an iterative manner which is guaranteed to converge. Initial instance selection is achieved by a simple yet effective kernel density estimator on the negative instances. Experimental results demonstrate the utility and efficiency of the proposed approach as compared to the state of the art.",
"Previous studies on multi-instance learning typically treated instances in the bags as independently and identically distributed. The instances in a bag, however, are rarely independent in real tasks, and a better performance can be expected if the instances are treated in an non-i.i.d. way that exploits relations among instances. In this paper, we propose two simple yet effective methods. In the first method, we explicitly map every bag to an undirected graph and design a graph kernel for distinguishing the positive and negative bags. In the second method, we implicitly construct graphs by deriving affinity matrices and propose an efficient graph kernel considering the clique information. The effectiveness of the proposed methods are validated by experiments.",
"The multiple instance problem arises in tasks where the training examples are ambiguous: a single example object may have many alternative feature vectors (instances) that describe it, and yet only one of those feature vectors may be responsible for the observed classification of the object. This paper describes and compares three kinds of algorithms that learn axis-parallel rectangles to solve the multiple instance problem. Algorithms that ignore the multiple instance problem perform very poorly. An algorithm that directly confronts the multiple instance problem (by attempting to identify which feature vectors are responsible for the observed classifications) performs best, giving 89 correct predictions on a musk odor prediction task. The paper also illustrates the use of artificial data to debug and compare these algorithms.",
"This paper presents two new formulations of multiple-instance learning as a maximum margin problem. The proposed extensions of the Support Vector Machine (SVM) learning approach lead to mixed integer quadratic programs that can be solved heuristic ally. Our generalization of SVMs makes a state-of-the-art classification technique, including non-linear classification via kernels, available to an area that up to now has been largely dominated by special purpose methods. We present experimental results on a pharmaceutical data set and on applications in automated image indexing and document categorization."
]
} |
1902.05066 | 2921001691 | Multi-instance learning (MIL) deals with tasks where data consist of set of bags and each bag is represented by a set of instances. Only the bag labels are observed but the label for each instance is not available. Previous MIL studies typically assume that the training and test samples follow the same distribution, which is often violated in real-world applications. Existing methods address distribution changes by re-weighting the training data with the density ratio between the training and test samples. However, models are often trained without prior knowledge of the test distribution which renders existing methods inapplicable. Inspired by a connection between MIL and causal inference, we propose a novel framework for addressing distribution change in MIL without relying on the test distribution. Experimental results validate the effectiveness of our approach. | Distribution robust supervised learning (DRSL) has been attracting research interest during the last few years @cite_3 @cite_8 . These methods minimize the empirical risk with regard to the worst case test distribution within a specific uncertainty set considered by the algorithm. Unfortunately, none of them considers the setting of multi-instance learning, and directly applying single instance distribution change methods to multi-instance learning is difficult with studies shown that trivial extensions do not improve the performances @cite_2 . | {
"cite_N": [
"@cite_2",
"@cite_3",
"@cite_8"
],
"mid": [
"2115672776",
"2108018401",
"2788557041"
],
"abstract": [
"The goal of traditional multi-instance learning (MIL) is to predict the labels of the bags, whereas in many real applications, it is desirable to get the instance labels, especially the labels of key instances that trigger the bag labels, in addition to getting bag labels. Such a problem has been largely unexplored before. In this paper, we formulate the Key Instance Detection (KID) problem, and propose a voting framework (VF) solution to KID. The key of VF is to exploit the relationship among instances, represented by a citer kNN graph. This graph is dierent from commonly used nearest neighbor graphs, but is suitable for KID. Experiments validate the eectiveness of VF for KID. Additionally, VF also outperforms state-of-the-art MIL approaches on the performance of bag label prediction.",
"In many important machine learning applications, the source distribution used to estimate a probabilistic classifier differs from the target distribution on which the classifier will be used to make predictions. Due to its asymptotic properties, sample reweighted empirical loss minimization is a commonly employed technique to deal with this difference. However, given finite amounts of labeled source data, this technique suffers from significant estimation errors in settings with large sample selection bias. We develop a framework for learning a robust bias-aware (RBA) probabilistic classifier that adapts to different sample selection biases using a minimax estimation formulation. Our approach requires only accurate estimates of statistics under the source distribution and is otherwise as robust as possible to unknown properties of the conditional label distribution, except when explicit generalization assumptions are incorporated. We demonstrate the behavior and effectiveness of our approach on binary classification tasks.",
""
]
} |
1902.05066 | 2921001691 | Multi-instance learning (MIL) deals with tasks where data consist of set of bags and each bag is represented by a set of instances. Only the bag labels are observed but the label for each instance is not available. Previous MIL studies typically assume that the training and test samples follow the same distribution, which is often violated in real-world applications. Existing methods address distribution changes by re-weighting the training data with the density ratio between the training and test samples. However, models are often trained without prior knowledge of the test distribution which renders existing methods inapplicable. Inspired by a connection between MIL and causal inference, we propose a novel framework for addressing distribution change in MIL without relying on the test distribution. Experimental results validate the effectiveness of our approach. | Several MIL algorithms have been proposed to address distribution change by utilizing the test distribution to calculate importance weight. MICS @cite_11 solves the problem by modeling the bag-level and instance-level distribution changes together. MIKI @cite_2 proposes that the distributions of positive instances should be modeled explicitly to address shifts of the positive concepts. | {
"cite_N": [
"@cite_2",
"@cite_11"
],
"mid": [
"2115672776",
"2181519955"
],
"abstract": [
"The goal of traditional multi-instance learning (MIL) is to predict the labels of the bags, whereas in many real applications, it is desirable to get the instance labels, especially the labels of key instances that trigger the bag labels, in addition to getting bag labels. Such a problem has been largely unexplored before. In this paper, we formulate the Key Instance Detection (KID) problem, and propose a voting framework (VF) solution to KID. The key of VF is to exploit the relationship among instances, represented by a citer kNN graph. This graph is dierent from commonly used nearest neighbor graphs, but is suitable for KID. Experiments validate the eectiveness of VF for KID. Additionally, VF also outperforms state-of-the-art MIL approaches on the performance of bag label prediction.",
"Multi-instance learning deals with tasks where each example is a bag of instances, and the bag labels of training data are known whereas instance labels are unknown. Most previous studies on multi-instance learning assumed that the training and testing data are from the same distribution; however, this assumption is often violated in real tasks. In this paper, we present possibly the first study on multi-instance learning with distribution change. We propose the MICS approach by considering both bag-level and instance-level distribution change. Experiments show that MICS is almost always significantly better than many state-of-theart multi-instance learning algorithms when distribution change occurs; and even when there is no distribution change, their performances are still comparable"
]
} |
1902.05135 | 2963805652 | Monitoring kernel object modification of virtual machine is widely used by virtual-machine-introspection-based security monitors to protect virtual machines in cloud computing, such as monitoring dentry objects to intercept file operations, etc. However, most of the current virtual machine monitors, such as KVM and Xen, only support page-level monitoring, because the Intel EPT technology can only monitor page privilege. If the out-of-virtual-machine security tools want to monitor some kernel objects, they need to intercept the operation of the whole memory page. Since there are some other objects stored in the monitored pages, the modification of them will also trigger the monitor. Therefore, page-level memory monitor usually introduces overhead to related kernel services of the target virtual machine. In this paper, we propose a low-overhead kernel object monitoring approach to reduce the overhead caused by page-level monitor. The core idea is to migrate the target kernel objects to a protected memory area and then to monitor the corresponding new memory pages. Since the new pages only contain the kernel objects to be monitored, other kernel objects will not trigger our monitor. Therefore, our monitor will not introduce runtime overhead to the related kernel service. The experimental results show that our system can monitor target kernel objects effectively only with very low overhead. | During the development of cloud computing, security is always a hot topic. Some related security issues are studied in @cite_22 @cite_8 @cite_2 @cite_16 . To protect VMs, Virtual Machine Introspection @cite_15 technology is proposed. It is an out-of-the-box technology to monitor VMs and is widely used in cloud security. In the virtualization architecture, VMM provides virtualization service for VMs , which has the highest privilege and is strongly isolated from the VMs. As a result, the security monitor running inside the VMM can access the VM hardware execution state. Based on VMI technology, many security monitor systems @cite_9 @cite_10 @cite_3 have been proposed to protect VMs. Compared with the in-VM security tools @cite_7 @cite_18 , VMI-based monitors are more secure and transparent. | {
"cite_N": [
"@cite_18",
"@cite_22",
"@cite_7",
"@cite_8",
"@cite_9",
"@cite_3",
"@cite_2",
"@cite_15",
"@cite_16",
"@cite_10"
],
"mid": [
"1971977519",
"",
"1595970205",
"2170018742",
"1999687070",
"2095519872",
"2080765376",
"1641762327",
"",
"2809988613"
],
"abstract": [
"An integrity checking and recovery (ICAR) system is presented here, which protects file system integrity and automatically restores modified files. The system enables files cryptographic hashes generation and verification, as well as configuration of security constraints. All of the crucial data, including ICAR system binaries, file backups and hashes database are stored in a physically write-protected storage to eliminate the threat of unauthorised modification. A buffering mechanism was designed and implemented in the system to increase operation performance. Additionally, the system supplies user tools for cryptographic hash generation and security database management. The system is implemented as a kernel extension, compliant with the Linux security model. Experimental evaluation of the system was performed and showed an approximate 10 performance degradation in secured file access compared to regular access.",
"",
"This paper contains an enhancement solution to an existing system called: Integrity Checking and Recovery (ICAR) system. ICAR provides a means to check for file integrity and also a feature to restore the file if a breach has been detected. The fact that it uses write protected storage makes it efficient against attacks. The performance degradation issue faced in ICAR system has been reduced here, so that users can now access secure files almost as normal as non-secure files. The improvement is achieved by making a modification to the cache present in the ICAR system. This is done by implementing the concept of augmentation of data structures in LRU page replacement algorithm thus making page operations faster in the cache which improves the overall performance of the system.",
"Abstract Security is critical for sensor networks used in military, homeland security and other hostile environments. Previous research on sensor network security mainly considers homogeneous sensor networks. Research has shown that homogeneous ad hoc networks have poor performance and scalability. Furthermore, many security schemes designed for homogeneous sensor networks suffer from high communication overhead, computation overhead, and or high storage requirement. Recently deployed sensor network systems are increasingly following heterogeneous designs. Key management is an essential cryptographic primitive to provide other security operations. In this paper, we present an effective key management scheme that takes advantage of the powerful high-end sensors in heterogeneous sensor networks. The performance evaluation and security analysis show that the key management scheme provides better security with low complexity and significant reduction on storage requirement, compared with existing key management schemes.",
"Over the past 20 years, we have witnessed a widespread adoption of dynamic binary instrumentation (DBI) for numerous program analyses and security applications including program debugging, profiling, reverse engineering, and malware analysis. To date, there are many DBI platforms, and the most popular one is Pin, which provides various instrumentation APIs for process instrumentation. However, Pin does not support the instrumentation of OS kernels. In addition, the execution of the instrumentation and analysis routine is always inside the virtual machine (VM). Consequently, it cannot support any out-of-VM introspection that requires strong isolation. Therefore, this paper presents PEMU, a new open source DBI framework that is compatible with Pin-APIs, but supports out-of-VM introspection for both user level processes and OS kernels. Unlike in-VM instrumentation in which there is no semantic gap, for out-of-VM introspection we have to bridge the semantic gap and provide abstractions (i.e., APIs) for programmers. One important feature of PEMU is its API compatibility with Pin. As such, many Pin plugins are able to execute atop PEMU without any source code modification. We have implemented PEMU, and our experimental results with the SPEC 2006 benchmarks show that PEMU introduces reasonable overhead.",
"The ability to introspect into the behavior of software at runtime is crucial for many security-related tasks, such as virtual machine-based intrusion detection and low-artifact malware analysis. Although some progress has been made in this task by automatically creating programs that can passively retrieve kernel-level information, two key challenges remain. First, it is currently difficult to extract useful information from user-level applications, such as web browsers. Second, discovering points within the OS and applications to hook for active monitoring is still an entirely manual process. In this paper we propose a set of techniques to mine the memory accesses made by an operating system and its applications to locate useful places to deploy active monitoring, which we call tap points. We demonstrate the efficacy of our techniques by finding tap points for useful introspection tasks such as finding SSL keys and monitoring web browser activity on five different operating systems (Windows 7, Linux, FreeBSD, Minix and Haiku) and two processor architectures (ARM and x86).",
"Wireless sensor networks have many applications, vary in size, and are deployed in a wide variety of areas. They are often deployed in potentially adverse or even hostile environment so that there are concerns on security issues in these networks. Sensor nodes used to form these networks are resource-constrained, which make security applications a challenging problem. Efficient key distribution and management mechanisms are needed besides lightweight ciphers. Many key establishment techniques have been designed to address the tradeoff between limited memory and security, but which scheme is the most effective is still debatable. In this paper, we provide a survey of key management schemes in wireless sensor networks. We notice that no key distribution technique is ideal to all the scenarios where sensor networks are used; therefore the techniques employed must depend upon the requirements of target applications and resources of each individual sensor network.",
"Today’s architectures for intrusion detection force the IDS designer to make a difficult choice. If the IDS resides on the host, it has an excellent view of what is happening in that host’s software, but is highly susceptible to attack. On the other hand, if the IDS resides in the network, it is more resistant to attack, but has a poor view of what is happening inside the host, making it more susceptible to evasion. In this paper we present an architecture that retains the visibility of a host-based IDS, but pulls the IDS outside of the host for greater attack resistance. We achieve this through the use of a virtual machine monitor. Using this approach allows us to isolate the IDS from the monitored host but still retain excellent visibility into the host’s state. The VMM also offers us the unique ability to completely mediate interactions between the host software and the underlying hardware. We present a detailed study of our architecture, including Livewire, a prototype implementation. We demonstrate Livewire by implementing a suite of simple intrusion detection policies and using them to detect real attacks.",
"",
"Sensitive files in computer systems such as executable programs, configuration, and authorization information have a great importance of their own, in terms of both confidentiality and functionality. To protect sensitive files, an effective approach named as file integrity monitoring is proposed to detect aggressive behaviors by verifying all the actions on these sensitive files. However, due to semantic gap problems, current file integrity monitoring tools are incapable of monitoring files in memory, so that an illegal modification of a file may bypass the detection by deliberately hiding itself inside the cache without actually committing to the disk. In this paper, we propose a runtime sensitive file integrity monitoring system named Vanguard, to satisfy the requirement of cache-level file protection. It can monitor both IO operations and cache operations, thereby deterring the illegal file accesses. To achieve the cache-level monitoring, we explore the techniques to detect when sensitive files are loaded into and swapped out from the operating system page cache. Vanguard is isolated from the monitored system so it is hard to be subverted. We implement Vanguard on QEMU KVM platform, and both Linux and Windows guest operating systems are supported. We conduct several experiments, and the experimental results show the effectiveness of Vanguard and imply that our method incurs acceptable overhead."
]
} |
1902.05135 | 2963805652 | Monitoring kernel object modification of virtual machine is widely used by virtual-machine-introspection-based security monitors to protect virtual machines in cloud computing, such as monitoring dentry objects to intercept file operations, etc. However, most of the current virtual machine monitors, such as KVM and Xen, only support page-level monitoring, because the Intel EPT technology can only monitor page privilege. If the out-of-virtual-machine security tools want to monitor some kernel objects, they need to intercept the operation of the whole memory page. Since there are some other objects stored in the monitored pages, the modification of them will also trigger the monitor. Therefore, page-level memory monitor usually introduces overhead to related kernel services of the target virtual machine. In this paper, we propose a low-overhead kernel object monitoring approach to reduce the overhead caused by page-level monitor. The core idea is to migrate the target kernel objects to a protected memory area and then to monitor the corresponding new memory pages. Since the new pages only contain the kernel objects to be monitored, other kernel objects will not trigger our monitor. Therefore, our monitor will not introduce runtime overhead to the related kernel service. The experimental results show that our system can monitor target kernel objects effectively only with very low overhead. | Many VMI-based security monitors @cite_14 @cite_19 @cite_11 enhance the VM security by intercepting the modification of VM kernel objects. CFWatcher @cite_14 is a real-time filesystem monitor which can intercept every access to the target files by monitoring the corresponding dentry or inode objects. RTKDSM @cite_11 enhances VM security by monitoring many kernel objects (such as FILE ) from outside. Because of the page-level intercepting, the extra overhead is introduced. | {
"cite_N": [
"@cite_19",
"@cite_14",
"@cite_11"
],
"mid": [
"2757620034",
"2498394137",
"1992496112"
],
"abstract": [
"",
"Protecting critical files in file systems is very important to computer systems. To protect critical files, the VMI-based Real-time File-system Monitor tools are promising options. However, these tools are always operation-based and introduce high overhead. The operation-based approaches intercept some kind of file operation to monitor critical files. The selected file operation is intercepted by the monitor whenever it is being executed. As file operation are high-frequency, the operationbased methods always result in the high performance degradation. In this paper, we present a VMI-based low overhead real-time critical file monitor method, CFWatcher, to meet the performance requirements of real-time monitor tools. CFWatcher is a target-based monitor tool which means it only intercepts the file operations accessing the user-defined critical files, and then obtains enough information to check the rules. The overhead of CFWatcher is related to the frequency of the target being accessed. Besides monitoring critical files, CFWatcher can take actions to prevent the illegal access if there is any rule violation. We implemented the prototype of CFWatcher and then evaluated the performance. Experimental results show that the overhead of our approach is low.",
"Virtual Machine Introspection (VMI) provides the ability to monitor virtual machines (VM) in an agentless fashion by gathering VM execution states from the hypervisor and analyzing those states to extract information about a running operating system (OS) without installing an agent inside the VM. VMI's main challenge lies in the difficulty in converting low-level byte string values into high-level semantic states of the monitored VM's OS. In this work, we tackle this challenge by developing a real-time kernel data structure monitoring (RTKDSM) system that leverages the rich OS analysis capabilities of Volatility, an open source computer forensics framework, to significantly simplify and automate analysis of VM execution states. The RTKDSM system is designed as an extensible software framework that is meant to be extended to perform application-specific VM state analysis. In addition, the RTKDSM system is able to perform real-time monitoring of any changes made to the extracted OS states of guest VMs. This real-time monitoring capability is especially important for VMI-based security applications. To minimize the performance overhead associated with real-time kernel data structure monitoring, the RTKDSM system has incorporated several optimizations whose effectiveness is reported in this paper."
]
} |
1902.05135 | 2963805652 | Monitoring kernel object modification of virtual machine is widely used by virtual-machine-introspection-based security monitors to protect virtual machines in cloud computing, such as monitoring dentry objects to intercept file operations, etc. However, most of the current virtual machine monitors, such as KVM and Xen, only support page-level monitoring, because the Intel EPT technology can only monitor page privilege. If the out-of-virtual-machine security tools want to monitor some kernel objects, they need to intercept the operation of the whole memory page. Since there are some other objects stored in the monitored pages, the modification of them will also trigger the monitor. Therefore, page-level memory monitor usually introduces overhead to related kernel services of the target virtual machine. In this paper, we propose a low-overhead kernel object monitoring approach to reduce the overhead caused by page-level monitor. The core idea is to migrate the target kernel objects to a protected memory area and then to monitor the corresponding new memory pages. Since the new pages only contain the kernel objects to be monitored, other kernel objects will not trigger our monitor. Therefore, our monitor will not introduce runtime overhead to the related kernel service. The experimental results show that our system can monitor target kernel objects effectively only with very low overhead. | To provide security service for cloud users, many systems have been proposed. CloudVMI @cite_13 provides the VMI interface for cloud customers, which can help them enhance the VM security. Furnace @cite_17 enables cloud customers to execute their VMI-based code in the VMM with high performance. Several papers (e.g., @cite_0 @cite_20 @cite_2 @cite_8 @cite_1 )have studied related issues. | {
"cite_N": [
"@cite_8",
"@cite_1",
"@cite_0",
"@cite_2",
"@cite_13",
"@cite_20",
"@cite_17"
],
"mid": [
"2170018742",
"1752315353",
"2162572433",
"2080765376",
"2022570654",
"2114337635",
"2892217134"
],
"abstract": [
"Abstract Security is critical for sensor networks used in military, homeland security and other hostile environments. Previous research on sensor network security mainly considers homogeneous sensor networks. Research has shown that homogeneous ad hoc networks have poor performance and scalability. Furthermore, many security schemes designed for homogeneous sensor networks suffer from high communication overhead, computation overhead, and or high storage requirement. Recently deployed sensor network systems are increasingly following heterogeneous designs. Key management is an essential cryptographic primitive to provide other security operations. In this paper, we present an effective key management scheme that takes advantage of the powerful high-end sensors in heterogeneous sensor networks. The performance evaluation and security analysis show that the key management scheme provides better security with low complexity and significant reduction on storage requirement, compared with existing key management schemes.",
"Recent advances in electronics and wireless communication technologies have enabled the development of large-scale wireless sensor networks that consist of many low-power, low-cost, and small-size sensor nodes. Sensor networks hold the promise of facilitating large-scale and real-time data processing in complex environments. Security is critical for many sensor network applications, such as military target tracking and security monitoring. To provide security and privacy to small sensor nodes is challenging, due to the limited capabilities of sensor nodes in terms of computation, communication, memory storage, and energy supply. In this article we survey the state of the art in research on sensor network security.",
"Internet protocol television (IPTV) will be the killer application for the next-generation Internet and will provide exciting new revenue opportunities for service providers. However, to deploy IPTV services with a full quality of service (QoS) guarantee, many underlying technologies must be further studied. This article serves as a survey of IPTV services and the underlying technologies. Technical challenges also are identified.",
"Wireless sensor networks have many applications, vary in size, and are deployed in a wide variety of areas. They are often deployed in potentially adverse or even hostile environment so that there are concerns on security issues in these networks. Sensor nodes used to form these networks are resource-constrained, which make security applications a challenging problem. Efficient key distribution and management mechanisms are needed besides lightweight ciphers. Many key establishment techniques have been designed to address the tradeoff between limited memory and security, but which scheme is the most effective is still debatable. In this paper, we provide a survey of key management schemes in wireless sensor networks. We notice that no key distribution technique is ideal to all the scenarios where sensor networks are used; therefore the techniques employed must depend upon the requirements of target applications and resources of each individual sensor network.",
"Virtual machine introspection (VMI) is a mechanism that allows indirect inspection and manipulation of the state of virtual machines. The indirection of this approach offers attractive isolation properties that has resulted in a variety of VMI-based applications dealing with security, performance, and debugging in virtual machine environments. Because it requires privileged access to the virtual machine monitor, VMI functionality is unfortunately not available to cloud users on public cloud platforms. In this paper, we present our work on the CloudVMI architecture to address this concern. CloudVMI virtualizes the VMI interface and makes it available as-a-service in a cloud environment. Because it allows introspection of users' VMs running on arbitrary physical machines in a cloud environment, our VMI-as-a-service abstraction allows a new class of cloud-centric VMI applications to be developed. We present the design and implementation of CloudVMI in the Xen hypervisor environment. We evaluate our implementation using a number of VMI applications, including a simple application that illustrates the cross-physical machine capabilities of CloudVMI.",
"Previous research on sensor network security mainly considers homogeneous sensor networks, where all sensor nodes have the same capabilities. Research has shown that homogeneous ad hoc networks have poor performance and scalability. The many-to-one traffic pattern dominates in sensor networks, and hence a sensor may only communicate with a small portion of its neighbors. Key management is a fundamental security operation. Most existing key management schemes try to establish shared keys for all pairs of neighbor sensors, no matter whether these nodes communicate with each other or not, and this causes large overhead. In this paper, we adopt a Heterogeneous Sensor Network (HSN) model for better performance and security. We propose a novel routing-driven key management scheme, which only establishes shared keys for neighbor sensors that communicate with each other. We utilize Elliptic Curve Cryptography in the design of an efficient key management scheme for sensor nodes. The performance evaluation and security analysis show that our key management scheme can provide better security with significant reductions on communication overhead, storage space and energy consumption than other key management schemes.",
"Although Virtual Machine Introspection (VMI) tools are increasingly capable, modern multi-tenant cloud providers are hesitant to expose the sensitive hypervisor APIs necessary for tenants to use them. Outside the cloud, VMI and virtualization-based security’s adoption rates are rising and increasingly considered necessary to counter sophisticated threats. This paper introduces Furnace, an open source VMI framework that outperforms prior frameworks by satisfying both a cloud provider’s expectation of security and a tenant’s desire to run their own custom VMI tools underneath their cloud VMs. Furnace’s flexibility and ease of use is demonstrated by porting four existing security and monitoring tools as Furnace VMI apps; these apps are shown to be resource efficient while executing up to 300x faster than those in previous VMI frameworks. Furnace’s security properties are shown to protect against the actions of malicious tenant apps."
]
} |
1902.05045 | 2911385740 | Recently, there has been a surge in interest in safe and robust techniques within reinforcement learning (RL). Current notions of risk in RL fail to capture the potential for systemic failures such as abrupt stoppages from system failures or surpassing of safety thresholds and the appropriate responsive controls in such instances. We propose a novel approach to risk minimisation within RL in which, in addition to taking actions that maximise its expected return, the controller learns a policy that is robust against stoppages due to an adverse event such as an abrupt failure. The results of the paper cover fault-tolerant control in worst-case scenarios under random stopping and optimal stopping, all in unknown environments. By demonstrating that the class of problems is represented by a variant of stochastic games, we prove the existence of a solution which is a unique fixed point equilibrium of the game and characterise the optimal controller behaviour. We then introduce a value function approximation algorithm that converges to the solution through simulation in unknown environments. | There is a plethora of work on OSPs in continuous and discrete-time . @cite_0 use approximate dynamic programming methods to construct an iterative scheme to compute the solution of an OSP. Our results generalise existing analyses to strategic settings with both a controller and an adversarial stopper which tackles risk within OSPs. OSPs have been extended to games where each player chooses one of two actions; to stop the game or continue. . | {
"cite_N": [
"@cite_0"
],
"mid": [
"2156168464"
],
"abstract": [
"The authors develop a theory characterizing optimal stopping times for discrete-time ergodic Markov processes with discounted rewards. The theory differs from prior work by its view of per-stage and terminal reward functions as elements of a certain Hilbert space. In addition to a streamlined analysis establishing existence and uniqueness of a solution to Bellman's equation, this approach provides an elegant framework for the study of approximate solutions. In particular, the authors propose a stochastic approximation algorithm that tunes weights of a linear combination of basis functions in order to approximate a value function. They prove that this algorithm converges (almost surely) and that the limit of convergence has some desirable properties. The utility of the approximation method is illustrated via a computational case study involving the pricing of a path dependent financial derivative security that gives rise to an optimal stopping problem with a 100-dimensional state space."
]
} |
1902.05026 | 2955961252 | There is growing interest in being able to run neural networks on sensors, wearables and internet-of-things (IoT) devices. However, the computational demands of neural networks make them difficult to deploy on resource-constrained edge devices. To meet this need, our work introduces a new recurrent unit architecture that is specifically adapted for on-device low power acoustic event detection (AED). The proposed architecture is based on the gated recurrent unit ('GRU' -- introduced by [9]) but features optimizations that make it implementable on ultra-low power micro-controllers such as the Arm Cortex M0+. Our new architecture, the Embedded Gated Recurrent Unit (eGRU) is demonstrated to be highly efficient and suitable for short-duration AED and keyword spotting tasks. A single eGRU cell is 60× faster and 10× smaller than a GRU cell. Despite its optimizations, eGRU compares well with GRU across tasks of varying complexities. The practicality of eGRU is investigated in a wearable acoustic event detection application. An eGRU model is implemented and tested on the Arm Cortex M0-based Atmel ATSAMD21E18 processor. The Arm M0+ implementation of the eGRU model compares favorably with a full precision GRU that is running on a workstation. The embedded eGRU model achieves a classification accuracy 95.3 , which is only 2 less than the full precision GRU. | Another common area of optimization is weight quantization. In previous works @cite_5 @cite_18 , 8-bit weight quantization was shown to drastically reduce network size without significant loss in accuracy. However, these approaches focused solely on convolutional and fully-connected networks. They also required the storage of a codebook for decoding weights at run-time. With regards to recurrent networks, @cite_31 demonstrated that 2-bit weight quantization is possible in GRU, with only a slight loss in accuracy. However, the combined effect of such a low precision quantization and other optimizations like a single gate architecture remained unknown. | {
"cite_N": [
"@cite_5",
"@cite_18",
"@cite_31"
],
"mid": [
"2119144962",
"2233116163",
"2512629640"
],
"abstract": [
"Neural networks are both computationally intensive and memory intensive, making them difficult to deploy on embedded systems with limited hardware resources. To address this limitation, we introduce \"deep compression\", a three stage pipeline: pruning, trained quantization and Huffman coding, that work together to reduce the storage requirement of neural networks by 35x to 49x without affecting their accuracy. Our method first prunes the network by learning only the important connections. Next, we quantize the weights to enforce weight sharing, finally, we apply Huffman coding. After the first two steps we retrain the network to fine tune the remaining connections and the quantized centroids. Pruning, reduces the number of connections by 9x to 13x; Quantization then reduces the number of bits that represent each connection from 32 to 5. On the ImageNet dataset, our method reduced the storage required by AlexNet by 35x, from 240MB to 6.9MB, without loss of accuracy. Our method reduced the size of VGG-16 by 49x from 552MB to 11.3MB, again with no loss of accuracy. This allows fitting the model into on-chip SRAM cache rather than off-chip DRAM memory. Our compression method also facilitates the use of complex neural networks in mobile applications where application size and download bandwidth are constrained. Benchmarked on CPU, GPU and mobile GPU, compressed network has 3x to 4x layerwise speedup and 3x to 7x better energy efficiency.",
"Recently, convolutional neural networks (CNN) have demonstrated impressive performance in various computer vision tasks. However, high performance hardware is typically indispensable for the application of CNN models due to the high computation complexity, which prohibits their further extensions. In this paper, we propose an efficient framework, namely Quantized CNN, to simultaneously speed-up the computation and reduce the storage and memory overhead of CNN models. Both filter kernels in convolutional layers and weighting matrices in fully-connected layers are quantized, aiming at minimizing the estimation error of each layer's response. Extensive experiments on the ILSVRC-12 benchmark demonstrate 4 6× speed-up and 15 20× compression with merely one percentage loss of classification accuracy. With our quantized CNN model, even mobile devices can accurately classify images within one second.",
"Recurrent Neural Networks (RNNs) produce state-of-art performance on many machine learning tasks but their demand on resources in terms of memory and computational power are often high. Therefore, there is a great interest in optimizing the computations performed with these models especially when considering development of specialized low-power hardware for deep networks. One way of reducing the computational needs is to limit the numerical precision of the network weights and biases. This has led to different proposed rounding methods which have been applied so far to only Convolutional Neural Networks and Fully-Connected Networks. This paper addresses the question of how to best reduce weight precision during training in the case of RNNs. We present results from the use of different stochastic and deterministic reduced precision training methods applied to three major RNN types which are then tested on several datasets. The results show that the weight binarization methods do not work with the RNNs. However, the stochastic and deterministic ternarization, and pow2-ternarization methods gave rise to low-precision RNNs that produce similar and even higher accuracy on certain datasets therefore providing a path towards training more efficient implementations of RNNs in specialized hardware."
]
} |
1902.05026 | 2955961252 | There is growing interest in being able to run neural networks on sensors, wearables and internet-of-things (IoT) devices. However, the computational demands of neural networks make them difficult to deploy on resource-constrained edge devices. To meet this need, our work introduces a new recurrent unit architecture that is specifically adapted for on-device low power acoustic event detection (AED). The proposed architecture is based on the gated recurrent unit ('GRU' -- introduced by [9]) but features optimizations that make it implementable on ultra-low power micro-controllers such as the Arm Cortex M0+. Our new architecture, the Embedded Gated Recurrent Unit (eGRU) is demonstrated to be highly efficient and suitable for short-duration AED and keyword spotting tasks. A single eGRU cell is 60× faster and 10× smaller than a GRU cell. Despite its optimizations, eGRU compares well with GRU across tasks of varying complexities. The practicality of eGRU is investigated in a wearable acoustic event detection application. An eGRU model is implemented and tested on the Arm Cortex M0-based Atmel ATSAMD21E18 processor. The Arm M0+ implementation of the eGRU model compares favorably with a full precision GRU that is running on a workstation. The embedded eGRU model achieves a classification accuracy 95.3 , which is only 2 less than the full precision GRU. | Using solely integer arithmetic in neural networks has also been studied in prior works @cite_5 @cite_2 @cite_9 . found that in a fully connected network, a dynamic point numeric format yields much better results than 20-bit fixed point. Another work demonstrates that 32-bit fixed point formats are effective in convolutional networks @cite_9 . Once again, all these efforts consider fully-connected and convolutional architectures. Since recurrent architectures are fundamentally different and much more difficult to train than other architectures @cite_23 , it is worthwhile to investigate fixed point arithmetic in recurrent neural networks. | {
"cite_N": [
"@cite_5",
"@cite_9",
"@cite_23",
"@cite_2"
],
"mid": [
"2119144962",
"2048266589",
"2949190276",
"1841592590"
],
"abstract": [
"Neural networks are both computationally intensive and memory intensive, making them difficult to deploy on embedded systems with limited hardware resources. To address this limitation, we introduce \"deep compression\", a three stage pipeline: pruning, trained quantization and Huffman coding, that work together to reduce the storage requirement of neural networks by 35x to 49x without affecting their accuracy. Our method first prunes the network by learning only the important connections. Next, we quantize the weights to enforce weight sharing, finally, we apply Huffman coding. After the first two steps we retrain the network to fine tune the remaining connections and the quantized centroids. Pruning, reduces the number of connections by 9x to 13x; Quantization then reduces the number of bits that represent each connection from 32 to 5. On the ImageNet dataset, our method reduced the storage required by AlexNet by 35x, from 240MB to 6.9MB, without loss of accuracy. Our method reduced the size of VGG-16 by 49x from 552MB to 11.3MB, again with no loss of accuracy. This allows fitting the model into on-chip SRAM cache rather than off-chip DRAM memory. Our compression method also facilitates the use of complex neural networks in mobile applications where application size and download bandwidth are constrained. Benchmarked on CPU, GPU and mobile GPU, compressed network has 3x to 4x layerwise speedup and 3x to 7x better energy efficiency.",
"Many companies are deploying services, either for consumers or industry, which are largely based on machine-learning algorithms for sophisticated processing of large amounts of data. The state-of-the-art and most popular such machine-learning algorithms are Convolutional and Deep Neural Networks (CNNs and DNNs), which are known to be both computationally and memory intensive. A number of neural network accelerators have been recently proposed which can offer high computational capacity area ratio, but which remain hampered by memory accesses. However, unlike the memory wall faced by processors on general-purpose workloads, the CNNs and DNNs memory footprint, while large, is not beyond the capability of the on-chip storage of a multi-chip system. This property, combined with the CNN DNN algorithmic characteristics, can lead to high internal bandwidth and low external communications, which can in turn enable high-degree parallelism at a reasonable area cost. In this article, we introduce a custom multi-chip machine-learning architecture along those lines. We show that, on a subset of the largest known neural network layers, it is possible to achieve a speedup of 450.65x over a GPU, and reduce the energy by 150.31x on average for a 64-chip system. We implement the node down to the place and route at 28nm, containing a combination of custom storage and computational units, with industry-grade interconnects.",
"There are two widely known issues with properly training Recurrent Neural Networks, the vanishing and the exploding gradient problems detailed in (1994). In this paper we attempt to improve the understanding of the underlying issues by exploring these problems from an analytical, a geometric and a dynamical systems perspective. Our analysis is used to justify a simple yet effective solution. We propose a gradient norm clipping strategy to deal with exploding gradients and a soft constraint for the vanishing gradients problem. We validate empirically our hypothesis and proposed solutions in the experimental section.",
"Training of large-scale deep neural networks is often constrained by the available computational resources. We study the effect of limited precision data representation and computation on neural network training. Within the context of low-precision fixed-point computations, we observe the rounding scheme to play a crucial role in determining the network's behavior during training. Our results show that deep networks can be trained using only 16-bit wide fixed-point number representation when using stochastic rounding, and incur little to no degradation in the classification accuracy. We also demonstrate an energy-efficient hardware accelerator that implements low-precision fixed-point arithmetic with stochastic rounding."
]
} |
1902.04570 | 2952014841 | In this paper, we introduce a variation of a state-of-the-art real-time tracker (CFNet), which adds to the original algorithm robustness to target loss without a significant computational overhead. The new method is based on the assumption that the feature map can be used to estimate the tracking confidence more accurately. When the confidence is low, we avoid updating the object's position through the feature map; instead, the tracker passes to a single-frame failure mode, during which the patch's low-level visual content is used to swiftly update the object's position, before recovering from the target loss in the next frame. The experimental evidence provided by evaluating the method on several tracking datasets validates both the theoretical assumption that the feature map is associated to tracking confidence, and that the proposed implementation can achieve target recovery in multiple scenarios, without compromising the real-time performance. | Due to its major significance, the tracking of moving objects in videos has a long history of research and development. The matureness of the domain partially explains the failure of deep-learning trackers to outperform classical'' trackers @cite_7 . Perhaps a more important reason is the inherent characteristics of the tracking setup that undermined a straightforward transfer of deep-learning techniques to this task @cite_18 . More specifically, (1) maximising heatmaps corresponding to semantic classes is not necessarily the optimal strategy to locate a specific object @cite_3 (especially in the presence of distractors), (2) off-line training is hampered by the lack of large-volume annotated datasets for video tracking, and (3) on-line training is prohibitively slow for applications requiring real-time tracking, especially if the model update is conducted in each and every frame @cite_9 @cite_12 . | {
"cite_N": [
"@cite_18",
"@cite_7",
"@cite_9",
"@cite_3",
"@cite_12"
],
"mid": [
"2584265939",
"1857884451",
"2557641257",
"2214352687",
"2770218765"
],
"abstract": [
"In this paper we introduce a fully end-to-end approach for visual tracking in videos that learns to predict the bounding box locations of a target object at every frame. An important insight is that the tracking problem can be considered as a sequential decision-making process and historical semantics encode highly relevant information for future decisions. Based on this intuition, we formulate our model as a recurrent convolutional neural network agent that interacts with a video overtime, and our model can be trained with reinforcement learning (RL) algorithms to learn good tracking policies that pay attention to continuous, inter-frame correlation and maximize tracking performance in the long run. The proposed tracking algorithm achieves state-of-the-art performance in an existing tracking benchmark and operates at frame-rates faster than real-time. To the best of our knowledge, our tracker is the first neural-network tracker that combines convolutional and recurrent networks with RL algorithms.",
"We propose a novel visual tracking algorithm based on the representations from a discriminatively trained Convolutional Neural Network (CNN). Our algorithm pretrains a CNN using a large set of videos with tracking groundtruths to obtain a generic target representation. Our network is composed of shared layers and multiple branches of domain-specific layers, where domains correspond to individual training sequences and each branch is responsible for binary classification to identify target in each domain. We train each domain in the network iteratively to obtain generic target representations in the shared layers. When tracking a target in a new sequence, we construct a new network by combining the shared layers in the pretrained CNN with a new binary classification layer, which is updated online. Online tracking is performed by evaluating the candidate windows randomly sampled around the previous target state. The proposed algorithm illustrates outstanding performance in existing tracking benchmarks.",
"In recent years, Discriminative Correlation Filter (DCF) based methods have significantly advanced the state-of-the-art in tracking. However, in the pursuit of ever increasing tracking performance, their characteristic speed and real-time capability have gradually faded. Further, the increasingly complex models, with massive number of trainable parameters, have introduced the risk of severe over-fitting. In this work, we tackle the key causes behind the problems of computational complexity and over-fitting, with the aim of simultaneously improving both speed and performance. We revisit the core DCF formulation and introduce: (i) a factorized convolution operator, which drastically reduces the number of parameters in the model, (ii) a compact generative model of the training sample distribution, that significantly reduces memory and time complexity, while providing better diversity of samples, (iii) a conservative model update strategy with improved robustness and reduced complexity. We perform comprehensive experiments on four benchmarks: VOT2016, UAV123, OTB-2015, and TempleColor. When using expensive deep features, our tracker provides a 20-fold speedup and achieves a 13.0 relative gain in Expected Average Overlap compared to the top ranked method [12] in the VOT2016 challenge. Moreover, our fast variant, using hand-crafted features, operates at 60 Hz on a single CPU, while obtaining 65.0 AUC on OTB-2015.",
"Visual object tracking is challenging as target objects often undergo significant appearance changes caused by deformation, abrupt motion, background clutter and occlusion. In this paper, we exploit features extracted from deep convolutional neural networks trained on object recognition datasets to improve tracking accuracy and robustness. The outputs of the last convolutional layers encode the semantic information of targets and such representations are robust to significant appearance variations. However, their spatial resolution is too coarse to precisely localize targets. In contrast, earlier convolutional layers provide more precise localization but are less invariant to appearance changes. We interpret the hierarchies of convolutional layers as a nonlinear counterpart of an image pyramid representation and exploit these multiple levels of abstraction for visual tracking. Specifically, we adaptively learn correlation filters on each convolutional layer to encode the target appearance. We hierarchically infer the maximum response of each layer to locate targets. Extensive experimental results on a largescale benchmark dataset show that the proposed algorithm performs favorably against state-of-the-art methods.",
"Long-term tracking requires extreme stability to the multitude of model updates and robustness to the disappearance and loss of the target as such will inevitably happen. For motivation, we have taken 10 randomly selected OTB-sequences, doubled each by attaching a reversed version and repeated each double sequence 20 times. On most of these repetitive videos, the best current tracker performs worse on each loop. This illustrates the difference between optimization for short-term versus long-term tracking. In a long-term tracker a combined global and local search strategy is beneficial, allowing for recovery from failures and disappearance. Most importantly, the proposed tracker also employs cautious updating, guided by self-quality assessment. The proposed tracker is still among the best on the 20-sec OTB-videos while achieving state-of-the-art on the 100-sec UAV20L benchmark. On 10 new half-an-hour videos with city bicycling, sport games etc, the proposed tracker outperforms others by a large margin where the 2010 TLD tracker comes second."
]
} |
1902.04570 | 2952014841 | In this paper, we introduce a variation of a state-of-the-art real-time tracker (CFNet), which adds to the original algorithm robustness to target loss without a significant computational overhead. The new method is based on the assumption that the feature map can be used to estimate the tracking confidence more accurately. When the confidence is low, we avoid updating the object's position through the feature map; instead, the tracker passes to a single-frame failure mode, during which the patch's low-level visual content is used to swiftly update the object's position, before recovering from the target loss in the next frame. The experimental evidence provided by evaluating the method on several tracking datasets validates both the theoretical assumption that the feature map is associated to tracking confidence, and that the proposed implementation can achieve target recovery in multiple scenarios, without compromising the real-time performance. | On the other hand, the small temporal window between two consecutive frames implies a visual similarity of the tracked object. Based on this rationale, a tracking-by-similarity tracker has been introduced @cite_8 , in which similarity is learned through a Siamese deep network that is trained offline, while the localisation is conducted through a correlation filter @cite_11 . This method achieved state-of-the-art real-time performance, despite its simple architecture. | {
"cite_N": [
"@cite_11",
"@cite_8"
],
"mid": [
"2962824803",
"2470394683"
],
"abstract": [
"The Correlation Filter is an algorithm that trains a linear template to discriminate between images and their translations. It is well suited to object tracking because its formulation in the Fourier domain provides a fast solution, enabling the detector to be re-trained once per frame. Previous works that use the Correlation Filter, however, have adopted features that were either manually designed or trained for a different task. This work is the first to overcome this limitation by interpreting the Correlation Filter learner, which has a closed-form solution, as a differentiable layer in a deep neural network. This enables learning deep features that are tightly coupled to the Correlation Filter. Experiments illustrate that our method has the important practical benefit of allowing lightweight architectures to achieve state-of-the-art performance at high framerates.",
"The problem of arbitrary object tracking has traditionally been tackled by learning a model of the object’s appearance exclusively online, using as sole training data the video itself. Despite the success of these methods, their online-only approach inherently limits the richness of the model they can learn. Recently, several attempts have been made to exploit the expressive power of deep convolutional networks. However, when the object to track is not known beforehand, it is necessary to perform Stochastic Gradient Descent online to adapt the weights of the network, severely compromising the speed of the system. In this paper we equip a basic tracking algorithm with a novel fully-convolutional Siamese network trained end-to-end on the ILSVRC15 dataset for object detection in video. Our tracker operates at frame-rates beyond real-time and, despite its extreme simplicity, achieves state-of-the-art performance in multiple benchmarks."
]
} |
1902.04570 | 2952014841 | In this paper, we introduce a variation of a state-of-the-art real-time tracker (CFNet), which adds to the original algorithm robustness to target loss without a significant computational overhead. The new method is based on the assumption that the feature map can be used to estimate the tracking confidence more accurately. When the confidence is low, we avoid updating the object's position through the feature map; instead, the tracker passes to a single-frame failure mode, during which the patch's low-level visual content is used to swiftly update the object's position, before recovering from the target loss in the next frame. The experimental evidence provided by evaluating the method on several tracking datasets validates both the theoretical assumption that the feature map is associated to tracking confidence, and that the proposed implementation can achieve target recovery in multiple scenarios, without compromising the real-time performance. | One of the main issues with such a tracker is the sampling drift @cite_12 , from which the tracker often fails to recover, that occurs in the case of abrupt object motion, camera motion, etc. Two main solutions have been recently proposed: (1) the heatmap generated in the final stage of the algorithm can incorporate the localisation of the ambiguity by not exhibiting a clear peak and in this case it rejects the position update @cite_5 , and (2) the model drift can be avoided through a global similarity search (in the whole frame) every @math frames ( @math being a hard-coded constant parameter) @cite_12 . | {
"cite_N": [
"@cite_5",
"@cite_12"
],
"mid": [
"2949213324",
"2770218765"
],
"abstract": [
"We introduce the OxUvA dataset and benchmark for evaluating single-object tracking algorithms. Benchmarks have enabled great strides in the field of object tracking by defining standardized evaluations on large sets of diverse videos. However, these works have focused exclusively on sequences that are just tens of seconds in length and in which the target is always visible. Consequently, most researchers have designed methods tailored to this \"short-term\" scenario, which is poorly representative of practitioners' needs. Aiming to address this disparity, we compile a long-term, large-scale tracking dataset of sequences with average length greater than two minutes and with frequent target object disappearance. The OxUvA dataset is much larger than the object tracking datasets of recent years: it comprises 366 sequences spanning 14 hours of video. We assess the performance of several algorithms, considering both the ability to locate the target and to determine whether it is present or absent. Our goal is to offer the community a large and diverse benchmark to enable the design and evaluation of tracking methods ready to be used \"in the wild\". The project website is this http URL",
"Long-term tracking requires extreme stability to the multitude of model updates and robustness to the disappearance and loss of the target as such will inevitably happen. For motivation, we have taken 10 randomly selected OTB-sequences, doubled each by attaching a reversed version and repeated each double sequence 20 times. On most of these repetitive videos, the best current tracker performs worse on each loop. This illustrates the difference between optimization for short-term versus long-term tracking. In a long-term tracker a combined global and local search strategy is beneficial, allowing for recovery from failures and disappearance. Most importantly, the proposed tracker also employs cautious updating, guided by self-quality assessment. The proposed tracker is still among the best on the 20-sec OTB-videos while achieving state-of-the-art on the 100-sec UAV20L benchmark. On 10 new half-an-hour videos with city bicycling, sport games etc, the proposed tracker outperforms others by a large margin where the 2010 TLD tracker comes second."
]
} |
1902.04570 | 2952014841 | In this paper, we introduce a variation of a state-of-the-art real-time tracker (CFNet), which adds to the original algorithm robustness to target loss without a significant computational overhead. The new method is based on the assumption that the feature map can be used to estimate the tracking confidence more accurately. When the confidence is low, we avoid updating the object's position through the feature map; instead, the tracker passes to a single-frame failure mode, during which the patch's low-level visual content is used to swiftly update the object's position, before recovering from the target loss in the next frame. The experimental evidence provided by evaluating the method on several tracking datasets validates both the theoretical assumption that the feature map is associated to tracking confidence, and that the proposed implementation can achieve target recovery in multiple scenarios, without compromising the real-time performance. | In this work, we carefully combine the above solutions in order to optimise both the accuracy and the computational time. The localisation ambiguity is detected using the Nearest Neighbour Distance Ratio (NNDR), instead of through a peak comparison to a hard-coded threshold. After rejecting an ambiguous position update, tracking is conducted in this frame using correlation of low-level patch representations as a backup tracker. This is inspired by the recent improvements on tracker performance that was achieved by including the first network layers, which model the low-level image content @cite_6 @cite_9 . Finally, in the next frame, the search window is expanded, but without covering the whole frame, as in @cite_12 , so as to reduce the computational cost. | {
"cite_N": [
"@cite_9",
"@cite_12",
"@cite_6"
],
"mid": [
"2557641257",
"2770218765",
"2518013266"
],
"abstract": [
"In recent years, Discriminative Correlation Filter (DCF) based methods have significantly advanced the state-of-the-art in tracking. However, in the pursuit of ever increasing tracking performance, their characteristic speed and real-time capability have gradually faded. Further, the increasingly complex models, with massive number of trainable parameters, have introduced the risk of severe over-fitting. In this work, we tackle the key causes behind the problems of computational complexity and over-fitting, with the aim of simultaneously improving both speed and performance. We revisit the core DCF formulation and introduce: (i) a factorized convolution operator, which drastically reduces the number of parameters in the model, (ii) a compact generative model of the training sample distribution, that significantly reduces memory and time complexity, while providing better diversity of samples, (iii) a conservative model update strategy with improved robustness and reduced complexity. We perform comprehensive experiments on four benchmarks: VOT2016, UAV123, OTB-2015, and TempleColor. When using expensive deep features, our tracker provides a 20-fold speedup and achieves a 13.0 relative gain in Expected Average Overlap compared to the top ranked method [12] in the VOT2016 challenge. Moreover, our fast variant, using hand-crafted features, operates at 60 Hz on a single CPU, while obtaining 65.0 AUC on OTB-2015.",
"Long-term tracking requires extreme stability to the multitude of model updates and robustness to the disappearance and loss of the target as such will inevitably happen. For motivation, we have taken 10 randomly selected OTB-sequences, doubled each by attaching a reversed version and repeated each double sequence 20 times. On most of these repetitive videos, the best current tracker performs worse on each loop. This illustrates the difference between optimization for short-term versus long-term tracking. In a long-term tracker a combined global and local search strategy is beneficial, allowing for recovery from failures and disappearance. Most importantly, the proposed tracker also employs cautious updating, guided by self-quality assessment. The proposed tracker is still among the best on the 20-sec OTB-videos while achieving state-of-the-art on the 100-sec UAV20L benchmark. On 10 new half-an-hour videos with city bicycling, sport games etc, the proposed tracker outperforms others by a large margin where the 2010 TLD tracker comes second.",
"Discriminative Correlation Filters (DCF) have demonstrated excellent performance for visual object tracking. The key to their success is the ability to efficiently exploit available negative data b ..."
]
} |
1902.04760 | 2913473169 | Several recent trends in machine learning theory and practice, from the design of state-of-the-art Gaussian Process to the convergence analysis of deep neural nets (DNNs) under stochastic gradient descent (SGD), have found it fruitful to study wide random neural networks. Central to these approaches are certain scaling limits of such networks. We unify these results by introducing a notion of a straightline that can express most neural network computations, and we characterize its scaling limit when its tensors are large and randomized. From our framework follows (1) the convergence of random neural networks to Gaussian processes for architectures such as recurrent neural networks, convolutional neural networks, residual networks, attention, and any combination thereof, with or without batch normalization; (2) conditions under which the -- that weights in backpropagation can be assumed to be independent from weights in the forward pass -- leads to correct computation of gradient dynamics, and corrections when it does not; (3) the convergence of the Neural Tangent Kernel, a recently proposed kernel used to predict training dynamics of neural networks under gradient descent, at initialization for all architectures in (1) without batch normalization. Mathematically, our framework is general enough to rederive classical random matrix results such as the semicircle and the Marchenko-Pastur laws, as well as recent results in neural network Jacobian singular values. We hope our work opens a way toward design of even stronger Gaussian Processes, initialization schemes to avoid gradient explosion vanishing, and deeper understanding of SGD dynamics in modern architectures. | @math Our technique is general enough to rederive the semicircle law for the Gaussian Orthogonal Ensemble and the Marchenko-Pastur Law for Wishart matrices @cite_10 . See warmup:semicirclelaw,warmup:marchenkopasturlaw . | {
"cite_N": [
"@cite_10"
],
"mid": [
"602904462"
],
"abstract": [
"The field of random matrix theory has seen an explosion of activity in recent years, with connections to many areas of mathematics and physics. However, this makes the current state of the field almost too large to survey in a single book. In this graduate text, we focus on one specific sector of the field, namely the spectral distribution of random Wigner matrix ensembles (such as the Gaussian Unitary Ensemble), as well as iid matrix ensembles. The text is largely self-contained and starts with a review of relevant aspects of probability theory and linear algebra. With over 200 exercises, the book is suitable as an introductory text for beginning graduate students seeking to enter the field."
]
} |
1902.04627 | 2748610804 | Testing cyber-physical systems involves the execution of test cases on target-machines equipped with the latest release of a software control system. When testing industrial robots, it is common that the target machines need to share some common resources, e.g., costly hardware devices, and so there is a need to schedule test case execution on the target machines, accounting for these shared resources. With a large number of such tests executed on a regular basis, this scheduling becomes difficult to manage manually. In fact, with manual test execution planning and scheduling, some robots may remain unoccupied for long periods of time and some test cases may not be executed. | Regression testing @cite_13 , i.e. the repeated testing of systems after changes were made, in covers a broad area of research works, including automatic test case generation @cite_12 , test suite prioritization and test suite reduction @cite_0 . There, the idea of controlling the time taken by optimization processes in test suite prioritization is not new @cite_7 . In test suite prioritization, @cite_18 proposed to use time-aware genetic algorithms to optimize the order in which to execute the test cases. Zhang further refined this approach in @cite_11 by using integer linear programming. On-demand test suite reduction @cite_36 also exploits integer linear programming for preserving the fault-detection capability of a test suite while performing test suite reduction. Cost-aware methods are also available for selecting minimal subsets of test cases covering a number of requirements @cite_17 @cite_16 . All these approaches participate in a general effort to better control the time allocated to the optimization algorithms when they are used in processes. Note however that test suite execution scheduling is different to prioritization or reduction as it deals with the notion of scheduling in time the execution of all test cases, without paying attention to any prioritization or reduction. | {
"cite_N": [
"@cite_18",
"@cite_7",
"@cite_36",
"@cite_17",
"@cite_0",
"@cite_16",
"@cite_13",
"@cite_12",
"@cite_11"
],
"mid": [
"",
"2136921053",
"2144927251",
"1978910953",
"1991292921",
"2094157810",
"2144600533",
"2100987397",
"2107940771"
],
"abstract": [
"",
"Regression testing is an expensive process used to validate modified software. Test case prioritization techniques improve the cost-effectiveness of regression testing by ordering test cases such that those that are more important are run earlier in the testing process. Many prioritization techniques have been proposed and evidence shows that they can be beneficial. It has been suggested, however, that the time constraints that can be imposed on regression testing by various software development processes can strongly affect the behavior of prioritization techniques. If this is correct, a better understanding of the effects of time constraints could lead to improved prioritization techniques and improved maintenance and testing processes. We therefore conducted a series of experiments to assess the effects of time constraints on the costs and benefits of prioritization techniques. Our first experiment manipulates time constraint levels and shows that time constraints do play a significant role in determining both the cost-effectiveness of prioritization and the relative cost-benefit trade-offs among techniques. Our second experiment replicates the first experiment, controlling for several threats to validity including numbers of faults present, and shows that the results generalize to this wider context. Our third experiment manipulates the number of faults present in programs to examine the effects of faultiness levels on prioritization and shows that faultiness level affects the relative cost-effectiveness of prioritization techniques. Taken together, these results have several implications for test engineers wishing to cost-effectively regression test their software systems. These include suggestions about when and when not to prioritize, what techniques to employ, and how differences in testing processes may relate to prioritization cost--effectiveness.",
"Most test suite reduction techniques aim to select, from a given test suite, a minimal representative subset of test cases that retains the same code coverage as the suite. Empirical studies have shown, however, that test suites reduced in this manner may lose fault detection capability. Techniques have been proposed to retain certain redundant test cases in the reduced test suite so as to reduce the loss in fault-detection capability, but these still do concede some degree of loss. Thus, these techniques may be applicable only in cases where loose demands are placed on the upper limit of loss in fault-detection capability. In this work we present an on-demand test suite reduction approach, which attempts to select a representative subset satisfying the same test requirements as an initial test suite conceding at most l loss in fault-detection capability for at least c of the instances in which it is applied. Our technique collects statistics about loss in fault-detection capability at the level of individual statements and models the problem of test suite reduction as an integer linear programming problem. We have evaluated our approach in the contexts of three scenarios in which it might be used. Our results show that most test suites reduced by our approach satisfy given fault detection capability demands, and that the approach compares favorably with an existing test suite reduction approach.",
"Context: In software development and maintenance, a software system may frequently be updated to meet rapidly changing user requirements. New test cases will be designed to ensure the correctness of new or modified functions, thus gradually increasing the test suite's size. Test suite reduction techniques aim to decrease the cost of regression testing by removing the redundant test cases from the test suite and then obtaining a representative set of test cases that still yield a high level of code coverage. Objective: Most of the existing reduction algorithms focus on decreasing the test suite's size. Yet, the differences in execution costs among test cases are usually significant and it may take a lot of execution time to run a test suite consisting of a few long-running test cases. This paper presents and empirically evaluates cost-aware algorithms that can produce the representative sets with lower execution costs. Method: We first use a cost-aware test case metric, called Irreplaceability, and its enhanced version, called EIrreplaceability, to evaluate the possibility that each test case can be replaced by others during test suite reduction. Furthermore, we construct a cost-aware framework that incorporates the concept of test irreplaceability into some well-known test suite reduction algorithms. Results: The effectiveness of the cost-aware framework is evaluated via the subject programs and test suites collected from the Software-artifact Infrastructure Repository - frequently chosen benchmarks for experimentally evaluating test suite reduction methods. The empirical results reveal that the presented algorithms produce representative sets that normally incur a low cost to yield a high level of test coverage. Conclusion: The presented techniques indeed enhance the capability of the traditional reduction algorithms to reduce the execution cost of a test suite. Especially for the additional Greedy algorithm, the presented techniques decrease the costs of the representative sets by 8.10-46.57 .",
"In continuous integration development environments, software engineers frequently integrate new or changed code with the mainline codebase. This can reduce the amount of code rework that is needed as systems evolve and speed up development time. While continuous integration processes traditionally require that extensive testing be performed following the actual submission of code to the codebase, it is also important to ensure that enough testing is performed prior to code submission to avoid breaking builds and delaying the fast feedback that makes continuous integration desirable. In this work, we present algorithms that make continuous integration processes more cost-effective. In an initial pre-submit phase of testing, developers specify modules to be tested, and we use regression test selection techniques to select a subset of the test suites for those modules that render that phase more cost-effective. In a subsequent post-submit phase of testing, where dependent modules as well as changed modules are tested, we use test case prioritization techniques to ensure that failures are reported more quickly. In both cases, the techniques we utilize are novel, involving algorithms that are relatively inexpensive and do not rely on code coverage information -- two requirements for conducting testing cost-effectively in this context. To evaluate our approach, we conducted an empirical study on a large data set from Google that we make publicly available. The results of our study show that our selection and prioritization techniques can each lead to cost-effectiveness improvements in the continuous integration process.",
"A trend in software testing is reducing the size of a test suite while preserving its overall quality. Given a test suite and a set of requirements covered by the suite, test suite reduction aims at selecting a subset of test cases that cover the same set of requirements. Even though this problem has received considerable attention, finding the smallest subset of test cases is still challenging and commonly-used approaches address this problem only with approximated solutions. When executing a single test case requires much manual effort (e.g., hours of preparation), finding the minimal subset is needed to reduce the testing costs. In this paper, we introduce a radically new approach to test suite reduction, called FLOWER, based on a search among network maximum flows. From a given test suite and the requirements covered by the suite, FLOWER forms a flow network (with specific constraints) that is then traversed to find its maximum flows. FLOWER leverages the Ford-Fulkerson method to compute maximum flows and Constraint Programming techniques to search among optimal flows. FLOWER is an exact method that computes a minimum-sized test suite, preserving the coverage of requirements. The experimental results show that FLOWER outperforms a non-optimized implementation of the Integer Linear Programming approach by 15-3000 times in terms of the time needed to find an optimal solution, and a simple greedy approach by 5-15 in terms of the size of reduced test suite.",
"When software is modified, during development and maintenance, it is regression tested to provide confidence that the changes did not introduce unexpected errors and that new features behave as expected. One important problem in regression testing is how to select a subset of test cases, from the test suite used for the original version of the software, when testing a modified version of the software. Regression-test-selection techniques address this problem. Safe regression-test-selection techniques select every test case in the test suite that may behave differently in the original and modified versions of the software. Among existing safe regression testing techniques, efficient techniques are often too imprecise and achieve little savings in testing effort, whereas precise techniques are too expensive when used on large systems. This paper presents a new regression-test-selection technique for Java programs that is safe, precise, and yet scales to large systems. It also presents a tool that implements the technique and studies performed on a set of subjects ranging from 70 to over 500 KLOC. The studies show that our technique can efficiently reduce the regression testing effort and, thus, achieve considerable savings.",
"In object oriented software development, automated unit test generation tools typically target one class at a time. A class, however, is usually part of a software project consisting of more than one class, and these are subject to changes over time. This context of a class offers significant potential to improve test generation for individual classes. In this paper, we introduce Continuous Test Generation (CTG), which includes automated unit test generation during continuous integration (i.e., infrastructure that regularly builds and tests software projects). CTG offers several benefits: First, it answers the question of how much time to spend on each class in a project. Second, it helps to decide in which order to test them. Finally, it answers the question of which classes should be subjected to test generation in the first place. We have implemented CTG using the EvoSuite unit test generation tool, and performed experiments using eight of the most popular open source projects available on GitHub, ten randomly selected projects from the SF100 corpus, and five industrial projects. Our experiments demonstrate improvements of up to +58 for branch coverage and up to +69 for thrown undeclared exceptions, while reducing the time spent on test generation by up to +83 .",
"Techniques for test-case prioritization re-order test cases to increase their rate of fault detection. When there is a fixed time budget that does not allow the execution of all the test cases, time-aware techniques for test-case prioritization may achieve a better rate of fault detection than traditional techniques for test-case prioritization. In this paper, we propose a novel approach to time-aware test-case prioritization using integer linear programming. To evaluate our approach, we performed experiments on two subject programs involving four techniques for our approach, two techniques for an approach to time-aware test-case prioritization based on genetic algorithms, and four traditional techniques for test-case prioritization. The empirical results indicate that two of our techniques outperform all the other techniques for the two subjects under the scenarios of both general and version-specific prioritization. The empirical results also indicate that some traditional techniques with lower analysis time cost for test-case prioritization may still perform competitively when the time budget is not quite tight."
]
} |
1902.04627 | 2748610804 | Testing cyber-physical systems involves the execution of test cases on target-machines equipped with the latest release of a software control system. When testing industrial robots, it is common that the target machines need to share some common resources, e.g., costly hardware devices, and so there is a need to schedule test case execution on the target machines, accounting for these shared resources. With a large number of such tests executed on a regular basis, this scheduling becomes difficult to manage manually. In fact, with manual test execution planning and scheduling, some robots may remain unoccupied for long periods of time and some test cases may not be executed. | Scheduling problems have been studied in other contexts for decades and an extensive body of research exists on resource-constrained approaches. The scheduling domain is divided into distinct areas such as process execution scheduling in operating systems and scheduling of workforces in a construction project. The scheduling problem of this paper belongs to a scheduling category named resource-constrained project scheduling problem (RCPSP; see @cite_3 @cite_14 @cite_20 for an extensive overview). RCPSP is concerned with finding schedules for resource-consuming tasks with precedence constraints in a fixed time horizon, such that the makespan is minimized @cite_20 . From the angle of RCPSP, global resources can be expressed as which are available with exactly one unit per timestep and can therefore only be consumed by a single job per timestep. | {
"cite_N": [
"@cite_14",
"@cite_20",
"@cite_3"
],
"mid": [
"",
"2140742285",
"2038345112"
],
"abstract": [
"",
"The resource-constrained project scheduling problem (RCPSP) consists of activities that must be scheduled subject to precedence and resource constraints such that the makespan is minimized. It has become a well-known standard problem in the context of project scheduling which has attracted numerous researchers who developed both exact and heuristic scheduling procedures. However, it is a rather basic model with assumptions that are too restrictive for many practical applications. Consequently, various extensions of the basic RCPSP have been developed. This paper gives an overview over these extensions. The extensions are classified according to the structure of the RCPSP. We summarize generalizations of the activity concept, of the precedence relations and of the resource constraints. Alternative objectives and approaches for scheduling multiple projects are discussed as well. In addition to popular variants and extensions such as multiple modes, minimal and maximal time lags, and net present value-based objectives, the paper also provides a survey of many less known concepts.",
"Abstract Project scheduling is concerned with single-item or small batch production where scarce resources have to be allocated to dependent activities over time. Applications can be found in diverse industries such as construction engineering, software development, etc. Also, project scheduling is increasingly important for make-to-order companies where the capacities have been cut down in order to meet lean management concepts. Likewise, project scheduling is very attractive for researchers, because the models in this area are rich and, hence, difficult to solve. For instance, the resource-constrained project scheduling problem contains the job shop scheduling problem as a special case. So far, no classification scheme exists which is compatible with what is commonly accepted in machine scheduling. Also, a variety of symbols are used by project scheduling researchers in order to denote one and the same subject. Hence, there is a gap between machine scheduling on the one hand and project scheduling on the other with respect to both, viz. a common notation and a classification scheme. As a matter of fact, in project scheduling, an ever growing number of papers is going to be published and it becomes more and more difficult for the scientific community to keep track of what is really new and relevant. One purpose of our paper is to close this gap. That is, we provide a classification scheme, i.e. a description of the resource environment, the activity characteristics, and the objective function, respectively, which is compatible with machine scheduling and which allows to classify the most important models dealt with so far. Also, we propose a unifying notation. The second purpose of this paper is to review some of the recent developments. More specifically, we review exact and heuristic algorithms for the single-mode and the multi-mode case, for the time–cost tradeoff problem, for problems with minimum and maximum time lags, for problems with other objectives than makespan minimization and, last but not least, for problems with stochastic activity durations."
]
} |
1902.04627 | 2748610804 | Testing cyber-physical systems involves the execution of test cases on target-machines equipped with the latest release of a software control system. When testing industrial robots, it is common that the target machines need to share some common resources, e.g., costly hardware devices, and so there is a need to schedule test case execution on the target machines, accounting for these shared resources. With a large number of such tests executed on a regular basis, this scheduling becomes difficult to manage manually. In fact, with manual test execution planning and scheduling, some robots may remain unoccupied for long periods of time and some test cases may not be executed. | RCPSP has been addressed by both exact methods @cite_32 @cite_15 @cite_4 @cite_27 , as well as heuristic methods @cite_29 @cite_21 . Due to the vast amount of literature, we will focus on CP OR-methods most closely related to the work of this paper. The clear trend in both CP and OR is to solve such problems with hybrid approaches, like, for instance, the work by @cite_34 or @cite_6 . Furthermore, , a subfamily of RCPSP addressing unary resources (in our terms global resources), have been effectively solved, e.g. by lazy clause generation @cite_24 . | {
"cite_N": [
"@cite_4",
"@cite_29",
"@cite_21",
"@cite_32",
"@cite_6",
"@cite_24",
"@cite_27",
"@cite_15",
"@cite_34"
],
"mid": [
"1527917700",
"2121852392",
"2070891082",
"2511017421",
"2027524005",
"2295480228",
"2125886152",
"1440160528",
"1505671383"
],
"abstract": [
"The Resource-constrained Project Scheduling Problem (Rcpsp), in which a schedule must obey the resource constraints and the precedence constraints between pairs of activities, is one of the most studied scheduling problems. An important variation of the problem (RcpspDc) is to find a schedule which maximises the net present value (discounted cash flow), when every activity has a given cash flow associated with it. Given the success of lazy clause generation (Lcg) approaches to solve Rcpsp with and without generalised precedence relations it seems worthwhile investigating Lcg's use on Rcpspdc. To do so, we must construct propagators for the net-present-value constraint that explain their propagation to the Lcg solver. In this paper we construct three different propagators for net-present-value constraints, and show how they can be used to rapidly solve RcpspDc.",
"Abstract We consider heuristic algorithms for the resource-constrained project scheduling problem. Starting with a literature survey, we summarize the basic components of heuristic approaches. We briefly describe so-called X -pass methods which are based on priority rules as well as metaheuristic algorithms. Subsequently, we present the results of our in-depth computational study. Here, we evaluate the performance of several state-of-the-art heuristics from the literature on the basis of a standard set of test instances and point out to the most promising procedures. Moreover, we analyze the behavior of the heuristics with respect to their components such as priority rules and metaheuristic strategy. Finally, we examine the impact of problem characteristics such as project size and resource scarceness on the performance.",
"This paper considers heuristics for the well-known resource-constrained project scheduling problem (RCPSP). It provides an update of our survey which was published in 2000. We summarize and categorize a large number of heuristics that have recently been proposed in the literature. Most of these heuristics are then evaluated in a computational study and compared on the basis of our standardized experimental design. Based on the computational results we discuss features of good heuristics. The paper closes with some remarks on our test design and a summary of the recent developments in research on heuristics for the RCPSP.",
"The resource-constrained project scheduling problem is a fundamental scheduling problem which comprises activities, scarce resources required by activities for their execution, and precedence relations between activities. The goal is to find an optimal schedule satisfying the resource and precedence constraints. These scheduling problems have many applications, ranging from production planning to project management. One of them concerns multi-modes of activities, in which each mode represents a different time-resource or resource-resource trade-off option. In recent years, constraint programming technologies with nogood learning have pushed the boundaries for exact solution methods on various resource-constrained scheduling problems, but, surprisingly, have not been applied on multi-mode resource-constrained project scheduling. In this paper, we investigate different constraint programming models and searches and show the superiority of such technologies in comparison to the current state of the art. Our best approach solved all remaining open instances from a well-established benchmark library.",
"Since their introduction, local search algorithms have consistently represented the state of the art in solution techniques for the classical job-shop scheduling problem. This dominance is despite the availability of powerful search and inference techniques for scheduling problems developed by the constraint programming community. In this paper, we introduce a simple hybrid algorithm for job-shop scheduling that leverages both the fast, broad search capabilities of modern tabu search algorithms and the scheduling-specific inference capabilities of constraint programming. The hybrid algorithm significantly improves the performance of a state-of-the-art tabu search algorithm for the job-shop problem and represents the first instance in which a constraint programming algorithm obtains performance competitive with the best local search algorithms. Furthermore, the variability in solution quality obtained by the hybrid is significantly lower than that of pure local search algorithms. Beyond performance demonstration, we perform a series of experiments that provide insights into the roles of the two component algorithms in the overall performance of the hybrid.",
"We revisit the standard hybrid CP SAT approach for solving disjunctive scheduling problems. Previous methods entail the creation of redundant clauses when lazily generating atoms standing for bounds modifications. We first describe an alternative method for handling lazily generated atoms without computational overhead. Next, we propose a novel conflict analysis scheme tailored for disjunctive scheduling. Our experiments on well known Job Shop Scheduling instances show compelling evidence of the efficiency of the learning mechanism that we propose. In particular this approach is very efficient for proving unfeasibility.",
"We present a generic exact method for minimizing the project duration of the resource-constrained project scheduling problem with generalized precedence relations (Rcpsp max). This is a very general scheduling model with applications areas such as project management and production planning. Our method uses lazy clause generation, i.e., a hybrid of finite domain and Boolean satisfiability solving, in order to apply no-good learning and conflict-driven search to the solution generation. Our experiments show the benefit of lazy clause generation for finding an optimal solution and proving its optimality in comparison to other state-of-the-art exact and non-exact methods. In comparison to other methods, our method is able to find better solutions faster on the Rcpsp max benchmarks. Indeed, our method closes 573 open problem instances and generates better solutions in most of the remaining instances. Surprisingly, although ours is an exact method, it outperforms the published non-exact methods on these benchmarks in terms of the quality of solutions.",
"Resource-constrained project scheduling with the objective of minimizing project duration (RCPSP) is one of the most studied scheduling problems. In this paper we consider the RCPSP with general temporal constraints and calendar constraints. Calendar constraints make some resources unavailable on certain days in the scheduling period and force activity execution to be delayed while resources are unavailable. They arise in practice from, e.g., unavailabilities of staff during public holidays and weekends. The resulting problems are challenging optimization problems. We develop not only four different constraint programming (CP) models to tackle the problem, but also a specialized propagator for the cumulative resource constraints taking the calendar constraints into account. This propagator includes the ability to explain its inferences so it can be used in a lazy clause generation solver. We compare these models, and different search strategies on a challenging set of benchmarks using a lazy clause generation solver. We close 83 of the open problems of the benchmark set, and show that CP solutions are highly competitive with existing Mip models of the problem.",
"The global cumulative constraint was proposed for modelling cumulative resources in scheduling problems for finite domain (FD) propagation. Since that time a great deal of research has investigated new stronger and faster filtering techniques for cumulative, but still most of these techniques only pay off in limited cases or are not scalable. Recently, the \"lazy clause generation\" hybrid solving approach has been devised which allows a finite domain propagation engine possible to take advantage of advanced SAT technology, by \"lazily\" creating a SAT model of an FD problem as computation progresses. This allows the solver to make use of SAT nogood learning and autonomous search capabilities. In this paper we show that using lazy clause generation where we model cumulative constraint by decomposition gives a very competitive implementation of cumulative resource problems. We are able to close a number of open problems from the well-established PSPlib benchmark library of resource-constrained project scheduling problems."
]
} |
1902.04627 | 2748610804 | Testing cyber-physical systems involves the execution of test cases on target-machines equipped with the latest release of a software control system. When testing industrial robots, it is common that the target machines need to share some common resources, e.g., costly hardware devices, and so there is a need to schedule test case execution on the target machines, accounting for these shared resources. With a large number of such tests executed on a regular basis, this scheduling becomes difficult to manage manually. In fact, with manual test execution planning and scheduling, some robots may remain unoccupied for long periods of time and some test cases may not be executed. | RCPSP is considered to be a generalization of where (JSS) is one of the best known @cite_8 . JSS is the special case of RCPSP where each operation uses exactly one resource, and FJSS () further extends JSS such that each operation can be processed on any machine from a given set. The FJSS is known to be NP-hard @cite_26 . | {
"cite_N": [
"@cite_26",
"@cite_8"
],
"mid": [
"21655972",
"1969848548"
],
"abstract": [
"The flexible job shop scheduling problem (FJSSP) is a generalization and extension of the classical job shop scheduling problem (JSSP) in which — prior to the sequencing of operations — an assignment of operations to machines is necessary. In this technical report, we are examining common instances for different specifications of the FJSSP. Furthermore, a new FJSSP specification is introduced, where similar machines are pooled to work centers, and the first and last work center are obligatory. For this practice-oriented problem specification new test instances are presented.",
"Resource-constrained project scheduling involves the scheduling of project activities subject to precedence and resource constraints in order to meet the objective(s) in the best possible way. The area covers a wide variety of problem types. The objective of this paper is to provide a survey of what we believe are the important recent developments in the area. Our main focus will be on the recent progress made in and the encouraging computational experience gained with the use of optimal solution procedures for the basic resource-constrained project scheduling problem (RCPSP) and important extensions. We illustrate how the branching rules, dominance and bounding arguments of a new depth- first branch-and-bound procedure can be extended to a rich variety of related problems: the generalized resource-constrained project scheduling problem, the resource-constrained project scheduling problem with generalized precedence relations, the preemptive resource-constrained project scheduling problem, the resource availability cost problem, and the resource-constrained project scheduling problem with various time resource(cost) trade-offs and discounted cash flows."
]
} |
1902.04627 | 2748610804 | Testing cyber-physical systems involves the execution of test cases on target-machines equipped with the latest release of a software control system. When testing industrial robots, it is common that the target machines need to share some common resources, e.g., costly hardware devices, and so there is a need to schedule test case execution on the target machines, accounting for these shared resources. With a large number of such tests executed on a regular basis, this scheduling becomes difficult to manage manually. In fact, with manual test execution planning and scheduling, some robots may remain unoccupied for long periods of time and some test cases may not be executed. | While OTS is closely related to FJSS, and efficient approaches to FJSS are known @cite_9 @cite_22 , there are some differences. First, in OTS, execution times are machine-independent. Second, each job in OTS consists of only one operation, while in FJSS one job can contain several operations, where there are precedences between the operations. Finally, some operations additionally require exclusive access to a global resource, preventing overlap with other operations. | {
"cite_N": [
"@cite_9",
"@cite_22"
],
"mid": [
"2046437335",
"181741320"
],
"abstract": [
"A hierarchical algorithm for the flexible job shop scheduling problem is described, based on the tabu search metaheuristic. Hierarchical strategies have been proposed in the literature for complex scheduling problems, and the tabu search metaheuristic, being able to cope with different memory levels, provides a natural background for the development of a hierarchical algorithm. For the case considered, a two level approach has been devised, based on the decomposition in a routing and a job shop scheduling subproblem, which is obtained by assigning each operation of each job to one among the equivalent machines. Both problems are tackled by tabu search. Coordination issues between the two hierarchical levels are considered. Unlike other hierarchical schemes, which are based on a one-way information flow, the one proposed here is based on a two-way information flow. This characteristic, together with the flexibility of local search strategies like tabu search, allows to adapt the same basic algorithm to different objective functions. Preliminary computational experience is reported.",
"Many scheduling problems involve reasoning about tasks which may or may not actually occur, so called optional tasks. The state-of-the-art approach to modelling and solving such problems makes use of interval variables which allow a start time of @math indicating the task does not run. In this paper we show we can model interval variables in a lazy clause generation solver, and create explaining propagators for scheduling constraints using these interval variables. Given the success of lazy clause generation on many scheduling problems, this combination appears to give a powerful new solving approach to scheduling problems with optional tasks. We demonstrate the new solving technology on well-studied flexible job-shop scheduling problems where we are able to close 36 open problems."
]
} |
1902.04705 | 2914053483 | Illuminant estimation plays a key role in digital camera pipeline system, it aims at reducing color casting effect due to the influence of non-white illuminant. Recent researches handle this task by using Convolution Neural Network (CNN) as a mapping function from input image to a single illumination vector. However, global mapping approaches are difficult to deal with scenes under multi-light-sources. In this paper, we proposed a self-adaptive single and multi-illuminant estimation framework, which includes the following novelties: (1) Learning local self-adaptive kernels from the entire image for illuminant estimation with encoder-decoder CNN structure; (2) Providing confidence measurement for the prediction; (3) Clustering-based iterative fitting for computing single and multi-illumination vectors. The proposed global-to-local aggregation is able to predict multi-illuminant regionally by utilizing global information instead of training in patches, as well as brings significant improvement for single illuminant estimation. We outperform the state-of-the-art methods on standard benchmarks with the largest relative improvement of 16 . In addition, we collect a dataset contains over 13k images for illuminant estimation and evaluation. The code and dataset is available on this https URL | There are many traditional methods try to determine the color of objects under the canonical illuminant with specific assumptions. Different assumptions have given rise to numerous illuminant estimation methods that can be divided into two main groups. First of these groups contains low-level statistic based methods, such as White-patch @cite_7 , Gray-world @cite_30 , Shades-of-Gray @cite_10 , Grey-Edge (1st and 2nd order) @cite_4 , using the bright pixels @cite_12 , . The second group includes machine-learning based methods, such as gamut mapping (pixel, edge, and intersection based) @cite_2 , natural image statistics @cite_11 , Bayesian learning @cite_3 , spatio-spectral learning (maximum likehood estimation) @cite_28 , using color edge moments @cite_18 , using regression trees with simple features from color distribution @cite_32 and performing various kinds of spatial localizations @cite_29 , @cite_26 . | {
"cite_N": [
"@cite_30",
"@cite_18",
"@cite_26",
"@cite_4",
"@cite_7",
"@cite_28",
"@cite_29",
"@cite_32",
"@cite_3",
"@cite_2",
"@cite_10",
"@cite_12",
"@cite_11"
],
"mid": [
"2052090926",
"",
"2949799648",
"",
"2177554204",
"2128059302",
"759139936",
"1945748771",
"2108303106",
"1925897490",
"1534426948",
"2287042286",
"2154264220"
],
"abstract": [
"A comprehensive mathematical model to account for colour constancy is formulated. Since the visual system is able to measure true object colour in complex scenes under a broad range of spectral compositions, for the illumination; it is assumed that the visual system must implicitly estimate and illuminant. The basic hypothesis is that the estimate of the illuminant is made on the basis of spatial information from the entire visual field. This estimate is then used by the visual system to arrive at an estimate of the (object) reflectance of the various subfields in the complex visual scene. The estimates are made by matching the inputs to the system to linear combinations of fixed bases and standards in the colour space. The model provides a general unified mathematical framework for related psychophysical phenomenology.",
"",
"We present Fast Fourier Color Constancy (FFCC), a color constancy algorithm which solves illuminant estimation by reducing it to a spatial localization task on a torus. By operating in the frequency domain, FFCC produces lower error rates than the previous state-of-the-art by 13-20 while being 250-3000 times faster. This unconventional approach introduces challenges regarding aliasing, directional statistics, and preconditioning, which we address. By producing a complete posterior distribution over illuminants instead of a single illuminant estimate, FFCC enables better training techniques, an effective temporal smoothing technique, and richer methods for error analysis. Our implementation of FFCC runs at 700 frames per second on a mobile device, allowing it to be used as an accurate, real-time, temporally-coherent automatic white balance algorithm.",
"",
"If color appearance is to be a useful feature in identifying an object, then color appearance must remain roughly constant when the object is viewed in different contexts. People maintain approximate color constancy despite variation in the color of nearby objects and despite variation in the spectral power distribution of the ambient light. Land's retinex algorithm is a model of human color constancy. We analyze the retinex algorithm and discuss its general properties. We show that the algorithm is too sensitive to changes in the color of nearby objects to serve as an adequate model of human color constancy.",
"We introduce an efficient maximum likelihood approach for one part of the color constancy problem: removing from an image the color cast caused by the spectral distribution of the dominating scene illuminant. We do this by developing a statistical model for the spatial distribution of colors in white balanced images (i.e., those that have no color cast), and then using this model to infer illumination parameters as those being most likely under our model. The key observation is that by applying spatial band-pass filters to color images one unveils color distributions that are unimodal, symmetric, and well represented by a simple parametric form. Once these distributions are fit to training data, they enable efficient maximum likelihood estimation of the dominant illuminant in a new image, and they can be combined with statistical prior information about the illuminant in a very natural manner. Experimental evaluation on standard data sets suggests that the approach performs well.",
"Color constancy is the problem of inferring the color of the light that illuminated a scene, usually so that the illumination color can be removed. Because this problem is underconstrained, it is often solved by modeling the statistical regularities of the colors of natural objects and illumination. In contrast, in this paper we reformulate the problem of color constancy as a 2D spatial localization task in a log-chrominance space, thereby allowing us to apply techniques from object detection and structured prediction to the color constancy problem. By directly learning how to discriminate between correctly white-balanced images and poorly white-balanced images, our model is able to improve performance on standard benchmarks by nearly 40 .",
"Illumination estimation is the process of determining the chromaticity of the illumination in an imaged scene in order to remove undesirable color casts through white-balancing. While computational color constancy is a well-studied topic in computer vision, it remains challenging due to the ill-posed nature of the problem. One class of techniques relies on low-level statistical information in the image color distribution and works under various assumptions (e.g. Grey-World, White-Patch, etc). These methods have an advantage that they are simple and fast, but often do not perform well. More recent state-of-the-art methods employ learning-based techniques that produce better results, but often rely on complex features and have long evaluation and training times. In this paper, we present a learning-based method based on four simple color features and show how to use this with an ensemble of regression trees to estimate the illumination. We demonstrate that our approach is not only faster than existing learning-based methods in terms of both evaluation and training time, but also gives the best results reported to date on modern color constancy data sets.",
"Computational color constancy is the task of estimating the true reflectances of visible surfaces in an image. In this paper we follow a line of research that assumes uniform illumination of a scene, and that the principal step in estimating reflectances is the estimation of the scene illuminant. We review recent approaches to illuminant estimation, firstly those based on formulae for normalisation of the reflectance distribution in an image - so-called grey-world algorithms, and those based on a Bayesian formulation of image formation. In evaluating these previous approaches we introduce a new tool in the form of a database of 568 high-quality, indoor and outdoor images, accurately labelled with illuminant, and preserved in their raw form, free of correction or normalisation. This has enabled us to establish several properties experimentally. Firstly automatic selection of grey-world algorithms according to image properties is not nearly so effective as has been thought. Secondly, it is shown that Bayesian illuminant estimation is significantly improved by the improved accuracy of priors for illuminant and reflectance that are obtained from the new dataset.",
"In his paper we introduce two improvements to the three-dimensional gamut mapping approach to computational colour constancy. This approach consist of two separate parts. First the possible solutions are constrained. This part is dependent on the diagonal model of illumination change, which in turn, is a function of the camera sensors. In this work we propose a robust method for relaxing this reliance on the diagonal model. The second part of the gamut mapping paradigm is to choose a solution from the feasible set. Currently there are two general approaches for doing so. We propose a hybrid method which embodies the benefits of both, and generally performs better than either. We provide results using both generated data and a carefully calibrated set of 321 images. In the case of the modification for diagonal model failure, we provide synthetic results using two cameras with a distinctly different degree of support for the diagonal model. Here we verify that the new method does indeed reduce error due to the diagonal model. We also verify that the new method for choosing the solution offers significant improvement, both in the case of synthetic data and with real images.",
"Colour constancy is a central problem for any visual system performing a task which requires stable perception of the colour world. To solve the colour constancy problem we estimate the colour of the prevailing light and then, at the second stage, remove it. Two of the most commonly used simple techniques for estimating the colour of the light are the Grey-World and Max-RGB algorithms. In this paper we begin by observing that this two colour constancy computations will respectively return the right answer if the average scene colour is grey or the maximum is white (and conversely, the degree of failure is proportional to the extent that these assumptions hold). We go on to ask the following question: “ Would we perform better colour constancy by assuming the scene average is some shade of grey?”. We give a mathematical answer to this question. Firstly, we show that Max-RGB and Grey-World are two instantia-tions of Minkowski norm. Secondly, that for a large cali-brated dataset L6 norm colour constancy works best over-all (we have improved the performance achieved by a sim-ple normalization based approach). Surprisingly we found performance to be similar to more elaborated algorithm.",
"",
"Existing color constancy methods are all based on specific assumptions such as the spatial and spectral characteristics of images. As a consequence, no algorithm can be considered as universal. However, with the large variety of available methods, the question is how to select the method that performs best for a specific image. To achieve selection and combining of color constancy algorithms, in this paper natural image statistics are used to identify the most important characteristics of color images. Then, based on these image characteristics, the proper color constancy algorithm (or best combination of algorithms) is selected for a specific image. To capture the image characteristics, the Weibull parameterization (e.g., grain size and contrast) is used. It is shown that the Weibull parameterization is related to the image attributes to which the used color constancy methods are sensitive. An MoG-classifier is used to learn the correlation and weighting between the Weibull-parameters and the image attributes (number of edges, amount of texture, and SNR). The output of the classifier is the selection of the best performing color constancy method for a certain image. Experimental results show a large improvement over state-of-the-art single algorithms. On a data set consisting of more than 11,000 images, an increase in color constancy performance up to 20 percent (median angular error) can be obtained compared to the best-performing single algorithm. Further, it is shown that for certain scene categories, one specific color constancy algorithm can be used instead of the classifier considering several algorithms."
]
} |
1902.04705 | 2914053483 | Illuminant estimation plays a key role in digital camera pipeline system, it aims at reducing color casting effect due to the influence of non-white illuminant. Recent researches handle this task by using Convolution Neural Network (CNN) as a mapping function from input image to a single illumination vector. However, global mapping approaches are difficult to deal with scenes under multi-light-sources. In this paper, we proposed a self-adaptive single and multi-illuminant estimation framework, which includes the following novelties: (1) Learning local self-adaptive kernels from the entire image for illuminant estimation with encoder-decoder CNN structure; (2) Providing confidence measurement for the prediction; (3) Clustering-based iterative fitting for computing single and multi-illumination vectors. The proposed global-to-local aggregation is able to predict multi-illuminant regionally by utilizing global information instead of training in patches, as well as brings significant improvement for single illuminant estimation. We outperform the state-of-the-art methods on standard benchmarks with the largest relative improvement of 16 . In addition, we collect a dataset contains over 13k images for illuminant estimation and evaluation. The code and dataset is available on this https URL | These assumptions based methods @cite_7 , @cite_30 , @cite_10 , @cite_4 , @cite_12 , , usually perform badly when there are no white patches in the scene. scene with large area of green grass or blue sea. The current industrial way of implementing AWB algorithms is based on combining statistic-based methods with feature learning. The approach mainly consists of: (1) scene recognition, (2) estimation under different scene @cite_31 . | {
"cite_N": [
"@cite_30",
"@cite_4",
"@cite_7",
"@cite_31",
"@cite_10",
"@cite_12"
],
"mid": [
"2052090926",
"",
"2177554204",
"1589420638",
"1534426948",
"2287042286"
],
"abstract": [
"A comprehensive mathematical model to account for colour constancy is formulated. Since the visual system is able to measure true object colour in complex scenes under a broad range of spectral compositions, for the illumination; it is assumed that the visual system must implicitly estimate and illuminant. The basic hypothesis is that the estimate of the illuminant is made on the basis of spatial information from the entire visual field. This estimate is then used by the visual system to arrive at an estimate of the (object) reflectance of the various subfields in the complex visual scene. The estimates are made by matching the inputs to the system to linear combinations of fixed bases and standards in the colour space. The model provides a general unified mathematical framework for related psychophysical phenomenology.",
"",
"If color appearance is to be a useful feature in identifying an object, then color appearance must remain roughly constant when the object is viewed in different contexts. People maintain approximate color constancy despite variation in the color of nearby objects and despite variation in the spectral power distribution of the ambient light. Land's retinex algorithm is a model of human color constancy. We analyze the retinex algorithm and discuss its general properties. We show that the algorithm is too sensitive to changes in the color of nearby objects to serve as an adequate model of human color constancy.",
"Scene recognition is extremely useful to improve different tasks involved in the Image Generation Pipeline of single sensor mobile devices (e.g., white balancing, autoexposure, etc). This demo showcases our scene recognition engine implemented on a Nokia N900 smartphone. The engine exploits an image representation directly obtainable in the IGP of mobile devices. The demo works in realtime and it is able to discriminate among different classes of scenes. The framework is built by employing the FCam API to have an easy and precise control of the mobile digital camera. Each acquired image (or frame of a video) is holistically represented starting from the statistics collected on DCT domain. This allow instant and \"free of charge\" feature extraction process since the DCT is always computed into the IGP of a mobile for storage purposes (i.e., JPEG or MPEG format). A SVM classifier is used to perform the final inference about the context of the scene.",
"Colour constancy is a central problem for any visual system performing a task which requires stable perception of the colour world. To solve the colour constancy problem we estimate the colour of the prevailing light and then, at the second stage, remove it. Two of the most commonly used simple techniques for estimating the colour of the light are the Grey-World and Max-RGB algorithms. In this paper we begin by observing that this two colour constancy computations will respectively return the right answer if the average scene colour is grey or the maximum is white (and conversely, the degree of failure is proportional to the extent that these assumptions hold). We go on to ask the following question: “ Would we perform better colour constancy by assuming the scene average is some shade of grey?”. We give a mathematical answer to this question. Firstly, we show that Max-RGB and Grey-World are two instantia-tions of Minkowski norm. Secondly, that for a large cali-brated dataset L6 norm colour constancy works best over-all (we have improved the performance achieved by a sim-ple normalization based approach). Surprisingly we found performance to be similar to more elaborated algorithm.",
""
]
} |
1902.04705 | 2914053483 | Illuminant estimation plays a key role in digital camera pipeline system, it aims at reducing color casting effect due to the influence of non-white illuminant. Recent researches handle this task by using Convolution Neural Network (CNN) as a mapping function from input image to a single illumination vector. However, global mapping approaches are difficult to deal with scenes under multi-light-sources. In this paper, we proposed a self-adaptive single and multi-illuminant estimation framework, which includes the following novelties: (1) Learning local self-adaptive kernels from the entire image for illuminant estimation with encoder-decoder CNN structure; (2) Providing confidence measurement for the prediction; (3) Clustering-based iterative fitting for computing single and multi-illumination vectors. The proposed global-to-local aggregation is able to predict multi-illuminant regionally by utilizing global information instead of training in patches, as well as brings significant improvement for single illuminant estimation. We outperform the state-of-the-art methods on standard benchmarks with the largest relative improvement of 16 . In addition, we collect a dataset contains over 13k images for illuminant estimation and evaluation. The code and dataset is available on this https URL | Other techniques like @cite_8 , @cite_33 , @cite_17 , @cite_14 , aim at using deep CNN to extract high level feature for illuminant estimation, and they consider CNN as a black box" global mapping function directly map the input image to its corresponding illumination vector. In this paper, we present a CNN architecture for generating a local-based self-adaptive meaningful kernel for its own input. The illumination vector will be calculated based on the refrence image computed by applying the learned kernel on its input image. | {
"cite_N": [
"@cite_14",
"@cite_17",
"@cite_33",
"@cite_8"
],
"mid": [
"2739489527",
"2519060067",
"2306965214",
"1960752652"
],
"abstract": [
"Improvements in color constancy have arisen from the use of convolutional neural networks (CNNs). However, the patch-based CNNs that exist for this problem are faced with the issue of estimation ambiguity, where a patch may contain insufficient information to establish a unique or even a limited possible range of illumination colors. Image patches with estimation ambiguity not only appear with great frequency in photographs, but also significantly degrade the quality of network training and inference. To overcome this problem, we present a fully convolutional network architecture in which patches throughout an image can carry different confidence weights according to the value they provide for color constancy estimation. These confidence weights are learned and applied within a novel pooling layer where the local estimates are merged into a global solution. With this formulation, the network is able to determine what to learn and how to pool automatically from color constancy datasets without additional supervision. The proposed network also allows for end-to-end training, and achieves higher efficiency and accuracy. On standard benchmarks, our network outperforms the previous state-of-the-art while achieving 120x greater efficiency.",
"Illuminant estimation to achieve color constancy is an ill-posed problem. Searching the large hypothesis space for an accurate illuminant estimation is hard due to the ambiguities of unknown reflections and local patch appearances. In this work, we propose a novel Deep Specialized Network (DS-Net) that is adaptive to diverse local regions for estimating robust local illuminants. This is achieved through a new convolutional network architecture with two interacting sub-networks, i.e. an hypotheses network (HypNet) and a selection network (SelNet). In particular, HypNet generates multiple illuminant hypotheses that inherently capture different modes of illuminants with its unique two-branch structure. SelNet then adaptively picks for confident estimations from these plausible hypotheses. Extensive experiments on the two largest color constancy benchmark datasets show that the proposed ‘hypothesis selection’ approach is effective to overcome erroneous estimation. Through the synergy of HypNet and SelNet, our approach outperforms state-of-the-art methods such as [1, 2, 3].",
"Computational color constancy aims to estimate the color of the light source. The performance of many vision tasks, such as object detection and scene understanding, may benefit from color constancy by estimating the correct object colors. Since traditional color constancy methods are based on specific assumptions, none of those methods can be used as a universal predictor. Further, shallow learning schemes are used for training-based color constancy approaches, suffering from limited learning capacity. In this paper, we propose a framework using Deep Neural Networks (DNNs) to obtain an accurate light source estimator to achieve color constancy. We formulate color constancy as a DNN-based regression approach to estimate the color of the light source. The model is trained using datasets of more than a million images. Experiments show that the proposed algorithm outperforms the state-of-the-art by 9 . Especially in cross dataset validation, reducing the median angular error by 35 . Further, in our implementation, the algorithm operates at more than @math fps during",
"In this work we describe a Convolutional Neural Network (CNN) to accurately predict the scene illumination. Taking image patches as input, the CNN works in the spatial domain without using hand-crafted features that are employed by most previous methods. The network consists of one convolutional layer with max pooling, one fully connected layer and three output nodes. Within the network structure, feature learning and regression are integrated into one optimization process, which leads to a more effective model for estimating scene illumination. This approach achieves state-of-the-art performance on a standard dataset of RAW images. Preliminary experiments on images with spatially varying illumination demonstrate the stability of the local illuminant estimation ability of our CNN."
]
} |
1902.04711 | 2914174770 | Recent studies highlighting the vulnerability of computer architecture to information leakage attacks have been a cause of significant concern. Among the various classes of microarchitectural attacks, cache timing channels are especially worrisome since they have the potential to compromise users' private data at high bit rates. Prior works have demonstrated the use of cache miss patterns to detect these attacks. We find that cache miss traces can be easily spoofed and thus they may not be able to identify smarter adversaries. In this work, we show that , which records the number of cache blocks owned by a specific process, can be leveraged as a stronger indicator for the presence of cache timing channels. We observe that the modulation of cache access latency in timing channels can be recognized through analyzing pairwise cache occupancy patterns. Our experimental results show that cache occupancy patterns cannot be easily obfuscated even by advanced adversaries that successfully evade cache miss-based detection. | Cache side- and covert timing channels have been demonstrated on real hardware in prior studies @cite_22 @cite_5 @cite_18 @cite_27 @cite_17 @cite_30 @cite_20 @cite_2 @cite_6 . | {
"cite_N": [
"@cite_30",
"@cite_18",
"@cite_22",
"@cite_6",
"@cite_27",
"@cite_2",
"@cite_5",
"@cite_20",
"@cite_17"
],
"mid": [
"",
"1427174644",
"2119028650",
"2901937773",
"2037322308",
"2775425142",
"1533145153",
"2795015774",
"1934458198"
],
"abstract": [
"",
"Sharing memory pages between non-trusting processes is a common method of reducing the memory footprint of multi-tenanted systems. In this paper we demonstrate that, due to a weakness in the Intel X86 processors, page sharing exposes processes to information leaks. We present FLUSH+RELOAD, a cache side-channel attack technique that exploits this weakness to monitor access to memory lines in shared pages. Unlike previous cache side-channel attacks, FLUSH+RELOAD targets the Last-Level Cache (i.e. L3 on processors with three cache levels). Consequently, the attack program and the victim do not need to share the execution core. We demonstrate the efficacy of the FLUSH+RELOAD attack by using it to extract the private encryption keys from a victim program running GnuPG 1.4.13. We tested the attack both between two unrelated processes in a single operating system and between processes running in separate virtual machines. On average, the attack is able to recover 96.7 of the bits of the secret key by observing a single signature or decryption round.",
"Third-party cloud computing represents the promise of outsourcing as applied to computation. Services, such as Microsoft's Azure and Amazon's EC2, allow users to instantiate virtual machines (VMs) on demand and thus purchase precisely the capacity they require when they require it. In turn, the use of virtualization allows third-party cloud providers to maximize the utilization of their sunk capital costs by multiplexing many customer VMs across a shared physical infrastructure. However, in this paper, we show that this approach can also introduce new vulnerabilities. Using the Amazon EC2 service as a case study, we show that it is possible to map the internal cloud infrastructure, identify where a particular target VM is likely to reside, and then instantiate new VMs until one is placed co-resident with the target. We explore how such placement can then be used to mount cross-VM side-channel attacks to extract information from a target VM on the same machine.",
"Information leakage of sensitive data has become one of the fast growing concerns among computer users. With adversaries turning to hardware for exploits, caches are frequently a target for timing channels since they present different timing profiles for cache miss and hit latencies. Such timing channels operate by having an adversary covertly communicate secrets to a spy simply through modulating resource timing without leaving any physical evidence. In this article, we demonstrate a new vulnerability exposed by cache coherence protocols where adversaries could manipulate the coherence states on certain cache blocks to alter cache access timing and communicate secrets illegitimately. Our threat model assumes the trojan and spy can either exploit explicitly shared read-only physical pages (e.g., shared library code), or use memory deduplication feature to implicitly force create shared physical pages. We demonstrate a template that adversaries may use to construct covert timing channels through manipulating combinations of coherence states and data placement in different caches. We investigate several classes of cache coherence protocols, and observe that both directory-based and snoopy protocols can be subject to covert timing channel attacks. We identify that the root cause of the vulnerability to be the existence of access latency difference for cache lines in read-only cache coherence states: Exlusive and Shared. For defense, we propose a slightly modified cache coherence scheme that will enable the last level cache to directly respond to read data requests in these read-only coherence states, and avoid any latency difference that could enable timing channels.",
"Recent exploration into the unique security challenges of cloud computing have shown that when virtual machines belonging to different customers share the same physical machine, new forms of cross-VM covert channel communication arise. In this paper, we explore one of these threats, L2 cache covert channels, and demonstrate the limits of these this threat by providing a quantification of the channel bit rates and an assessment of its ability to do harm. Through progressively refining models of cross-VM covert channels from the derived maximums, to implementable channels in the lab, and finally in Amazon EC2 itself we show how a variety of factors impact our ability to create effective channels. While we demonstrate a covert channel with considerably higher bit rate than previously reported, we assess that even at such improved rates, the harm of data exfiltration from these channels is still limited to the sharing of small, if important, secrets such as private keys.",
"Covert channels provide a secret communication medium between two malicious processes to exfiltrate information stealthily that violates the security policy of a system. In this paper, we demonstrate a new covert timing channel attack that exploits the CPU operating frequencies with different power governors in real system environment. In particular, we establish how two colluding processes-a trojan and a spy can modulate the CPU frequency to create a powerful, high-capacity and robust covert channel. We implement this covert channel both in a single threaded and simultaneous multi-threading (SMT) environment and show the feasibility of such a communication. Our experiments on Intel Xeon server platform demonstrate dynamic frequency scaling covert channels that can achieve up to 20 bits second.",
"Information security and privacy in general are major concerns that impede enterprise adaptation of shared or public cloud computing. Specifically, the concern of virtual machine (VM) physical co-residency stems from the threat that hostile tenants can leverage various forms of side channels (such as cache covert channels) to exfiltrate sensitive information of victims on the same physical system. However, on virtualized ×86 systems, covert channel attacks have not yet proven to be practical, and thus the threat is widely considered a \"potential risk\". In this paper, we present a novel covert channel attack that is capable of high-bandwidth and reliable data transmission in the cloud. We first study the application of existing cache channel techniques in a virtualized environment, and uncover their major insufficiency and difficulties. We then overcome these obstacles by (1) redesigning a pure timing-based data transmission scheme, and (2) exploiting the memory bus as a high-bandwidth covert channel medium. We further design and implement a robust communication protocol, and demonstrate realistic covert channel attacks on various virtualized ×86 systems. Our experiments show that covert channels do pose serious threats to information security in the cloud. Finally, we discuss our insights on covert channel mitigation in virtualized environments.",
"Most commercial multi-core processors incorporate hardware coherence protocols to support efficient data transfers and updates between their constituent cores. While hardware coherence protocols provide immense benefits for application performance by removing the burden of software-based coherence, we note that understanding the security vulnerabilities posed by such oft-used, widely-adopted processor features is critical for secure processor designs in the future. In this paper, we demonstrate a new vulnerability exposed by cache coherence protocol states. We present novel insights into how adversaries could cleverly manipulate the coherence states on shared cache blocks, and construct covert timing channels to illegitimately communicate secrets to the spy. We demonstrate 6 different practical scenarios for covert timing channel construction. In contrast to prior works, we assume a broader adversary model where the trojan and spy can either exploit explicitly shared read-only physical pages (e.g., shared library code), or use memory deduplication feature to implicitly force create shared physical pages. We demonstrate how adversaries can manipulate combinations of coherence states and data placement in different caches to construct timing channels. We also explore how adversaries could exploit multiple caches and their associated coherence states to improve transmission bandwidth with symbols encoding multiple bits. Our experimental results on commercial systems show that the peak transmission bandwidths of these covert timing channels can vary between 700 to 1100 Kbits sec. To the best of our knowledge, our study is the first to highlight the vulnerability of hardware cache coherence protocols to timing channels that can help computer architects to craft effective defenses against exploits on such critical processor features.",
"We present an effective implementation of the Prime Probe side-channel attack against the last-level cache. We measure the capacity of the covert channel the attack creates and demonstrate a cross-core, cross-VM attack on multiple versions of GnuPG. Our technique achieves a high attack resolution without relying on weaknesses in the OS or virtual machine monitor or on sharing memory between attacker and victim."
]
} |
1902.04711 | 2914174770 | Recent studies highlighting the vulnerability of computer architecture to information leakage attacks have been a cause of significant concern. Among the various classes of microarchitectural attacks, cache timing channels are especially worrisome since they have the potential to compromise users' private data at high bit rates. Prior works have demonstrated the use of cache miss patterns to detect these attacks. We find that cache miss traces can be easily spoofed and thus they may not be able to identify smarter adversaries. In this work, we show that , which records the number of cache blocks owned by a specific process, can be leveraged as a stronger indicator for the presence of cache timing channels. We observe that the modulation of cache access latency in timing channels can be recognized through analyzing pairwise cache occupancy patterns. Our experimental results show that cache occupancy patterns cannot be easily obfuscated even by advanced adversaries that successfully evade cache miss-based detection. | @cite_9 propose a generic framework that detects cache covert timing channel using correlations between cache conflict misses. ReplayConfusion @cite_16 records and replays cache access traces from the trojan and spy, and detects cache timing attacks based on differences of cache misses under different cache slice hashing functions. @cite_3 have demonstrated novel use of hardware prefetchers to stop cache timing channels. These techniques require customized hardware for either fine-grained conflict miss tracking or cache design changes. | {
"cite_N": [
"@cite_9",
"@cite_16",
"@cite_3"
],
"mid": [
"2540054680",
"2561988158",
"2808687865"
],
"abstract": [
"Information security and data privacy have steadily grown into major concerns in computing, especially given the rapid transition into the digital age for all needs--from healthcare to national defense. Among the many forms of information leakage, covert timing channels can be dangerous primarily because they involve two parties intentionally colluding to exfiltrate sensitive data by subverting the underlying system security policy. The attackers establish an illegitimate communication channel between two processes and transmit information via resource timing modulation, which does not leave any physical activity trace for later forensic analysis. Recent studies have shown the vulnerability of many popular computing environments, such as cloud computing, to these covert timing channels. With the advancements in software confinement mechanisms, shared processor hardware structures will be natural targets for malicious attackers to exploit and implement their covert-timing-based channels. In this work, the authors present a microarchitecture-level framework that detects the possible presence of covert timing channels on shared hardware. Their experiments demonstrate their ability to successfully detect different types of covert timing channels on various hardware structures and communication patterns.",
"Cache-based covert channel attacks use highly-tuned shared-cache conflict misses to pass information from a trojan to a spy process. Detecting such attacks is very challenging. State of the art detection mechanisms do not consider the general characteristics of such attacks and, instead, focus on specific communication protocols. As a result, they fail to detect attacks using different protocols and, hence, have limited coverage. In this paper, we make the following observation about these attacks: not only are the malicious accesses highly tuned to the mapping of addresses to the caches; they also follow a distinctive cadence as bits are being received. Changing the mapping of addresses to the caches substantially disrupts the conflict miss patterns, but retains the cadence. This is in contrast to benign programs. Based on this observation, we propose a novel, high-coverage approach to detect cache-based covert channel attacks. It is called ReplayConfusion, and is based on Record and deterministic Replay (RnR). After a program's execution is recorded, it is deterministically replayed using a different mapping of addresses to the caches. We then analyze the difference between the cache miss rate timelines of the two runs. If the difference function is both sizable and exhibits a periodic pattern, it indicates that there is an attack. This paper also introduces a new taxonomy of cache-based covert channel attacks, and shows that ReplayConfusion uncovers examples from all the categories. Finally, ReplayConfusion only needs simple hardware.",
"Cache timing channels are a form of information leakage that operate through modulating cache access latencies and ultimately exfiltrate sensitive user information to adversaries. Among the many forms of timing channels, covert channels are particularly dangerous as they involve two insider processes (trojan and spy) colluding with each other to send out sensitive information, and are often difficult to detect or prevent. In this paper, we propose Prefetch-guard, an efficient and low-cost mitigation mechanism against cache-based timing channels. Prefetch-guard leverages hardware prefetchers to obfuscate the effect of timing modulation intentionally created by the trojan and spy. Our detection mechanism identifies the target cache sets that are being exploited for information leakage, and cache blocks are prefetched to fuzz the pattern of cache misses and hits created to construct timing channel between the trojan and the spy. With prefetch-guard, we observe that the cache timing channels suffer a 53 bit error rate which makes it very hard or impossible for the spy to decipher any useful information."
]
} |
1902.04711 | 2914174770 | Recent studies highlighting the vulnerability of computer architecture to information leakage attacks have been a cause of significant concern. Among the various classes of microarchitectural attacks, cache timing channels are especially worrisome since they have the potential to compromise users' private data at high bit rates. Prior works have demonstrated the use of cache miss patterns to detect these attacks. We find that cache miss traces can be easily spoofed and thus they may not be able to identify smarter adversaries. In this work, we show that , which records the number of cache blocks owned by a specific process, can be leveraged as a stronger indicator for the presence of cache timing channels. We observe that the modulation of cache access latency in timing channels can be recognized through analyzing pairwise cache occupancy patterns. Our experimental results show that cache occupancy patterns cannot be easily obfuscated even by advanced adversaries that successfully evade cache miss-based detection. | Additionally, propose a machine learning-based approach that identifies cache side channels based on performance counters @cite_4 . We note that as a stronger indicator, cache occupancy observations can be leveraged together for more robust detection. Meanwhile, our detection framework using cache occupancy can be potentially incorporated with existing prevention techniques such as cache partitioning @cite_25 @cite_19 to build end-to-end cache timing channel defense solutions. In addition to improving cache resiliency, we note that other mechanisms to improve software robustness @cite_8 @cite_29 @cite_13 @cite_11 @cite_1 @cite_24 and memory reliability @cite_26 @cite_14 will improve overall system security. | {
"cite_N": [
"@cite_26",
"@cite_4",
"@cite_14",
"@cite_8",
"@cite_29",
"@cite_1",
"@cite_24",
"@cite_19",
"@cite_13",
"@cite_25",
"@cite_11"
],
"mid": [
"2054161218",
"2166844173",
"2015772022",
"2611422138",
"2490096331",
"2102960656",
"2750625138",
"",
"1965071495",
"2406174928",
"2023582707"
],
"abstract": [
"As main memory systems begin to face the scaling challenges from DRAM technology, future computer systems need to adapt to the emerging memory technologies like Phase-Change Memory (PCM or PRAM). While these newer technologies offer advantages such as storage density, non-volatility, and low energy consumption, they are constrained by limited write endurance that becomes more pronounced with process variation. In this paper, we propose a novel PRAM-based main memory system, RePRAM (Recycling PRAM), which leverages a group of faulty pages and recycles them in a managed way to significantly extend the PRAM lifetime while minimizing the performance impact. In particular, we explore two different dimensions of dynamic redundancy levels and group sizes, and design low-cost hardware and software support for RePRAM. Our proposed scheme involves minimal hardware modifications (that have less than 1 on-chip and off-chip area overheads). Also, our schemes can improve the PRAM lifetime by up to 43× (times) over a chip with no error correction capabilities, and outperform prior schemes such as DRM and ECP at a small fraction of the hardware cost. The performance overhead resulting from our scheme is less than 7 on average across 21 applications from SPEC2006, Splash-2, and PARSEC benchmark suites.",
"The proliferation of computers in any domain is followed by the proliferation of malware in that domain. Systems, including the latest mobile platforms, are laden with viruses, rootkits, spyware, adware and other classes of malware. Despite the existence of anti-virus software, malware threats persist and are growing as there exist a myriad of ways to subvert anti-virus (AV) software. In fact, attackers today exploit bugs in the AV software to break into systems. In this paper, we examine the feasibility of building a malware detector in hardware using existing performance counters. We find that data from performance counters can be used to identify malware and that our detection techniques are robust to minor variations in malware programs. As a result, after examining a small set of variations within a family of malware on Android ARM and Intel Linux platforms, we can detect many variations within that family. Further, our proposed hardware modifications allow the malware detector to run securely beneath the system software, thus setting the stage for AV implementations that are simpler and less buggy than software AV. Combined, the robustness and security of hardware AV techniques have the potential to advance state-of-the-art online malware detection.",
"DRAM technology challenges have increased the necessity to adapt to the emerging memory technologies like Phase-Change Memory (PCM or PRAM). While such emerging technologies provide benefits like storage density, nonvolatility, and low energy consumption, they are constrained by limited write endurance that becomes more pronounced with process variation. In this article, we explore a novel PRAM-based main memory system which resuscitates a group of faulty pages in a cost-effective manner to significantly extend the PCM main memory lifetime while minimizing the performance impact. In particular, we explore three different dimensions of dynamic redundancy levels and group sizes, and design low-cost hardware and software support for our proposed schemes. We aim to have minimal hardware modifications (that have less than 1p on-chip and off-chip area overheads). Also, our schemes can improve the PRAM lifetime by up to 105× (times) over a chip with no error correction capabilities, and outperform prior schemes such as DRM and ECP at a small fraction of the hardware cost. The performance overhead resulting from our scheme is less than 8p on average across 21 applications from SPEC2006, Splash-2, and PARSEC benchmark suites.",
"Unsafe memory accesses in programs written using popular programming languages like C and C++ have been among the leading causes of software vulnerability. Memory safety checkers, such as Softbound, enforce memory spatial safety by checking if accesses to array elements are within the corresponding array bounds. However, such checks often result in high execution time overhead due to the cost of executing the instructions associated with the bound checks. To mitigate this problem, techniques to eliminate redundant bound checks are needed. In this paper, we propose a novel framework, SIMBER, to eliminate redundant memory bound checks via statistical inference. In contrast to the existing techniques that primarily rely on static code analysis, our solution leverages a simple, model-based inference to identify redundant bound checks based on runtime statistics from past program executions. We construct a knowledge base containing sufficient conditions using variables inside functions, which are then applied adaptively to avoid future redundant checks at a function-level granularity. Our experimental results on real-world applications show that SIMBER achieves zero false positives. Also, our approach reduces the performance overhead by up to 86.94 over Softbound, and incurs a modest 1.7 code size increase on average to circumvent the redundant bound checks inserted by Softbound.",
"This paper presents a semantics-aware rule recommendation and enforcement (SARRE) system for taming information leakage on Android. SARRE leverages statistical analysis and a novel application of minimum path cover algorithm to identify system event paths from dynamic runtime monitoring. Then, an online recommendation system is developed to automatically assign a fine-grained security rule to each event path, capitalizing on both known security rules and application semantic information. The proposed SARRE system is prototyped on Android devices and evaluated using real-world malware samples and popular apps from Google Play spanning multiple categories. Our results show that SARRE achieves 93.8 precision and 96.4 recall in identifying the event paths, compared with tainting technique. Also, the average difference between rule recommendation and manual configuration is less than 5 , validating the effectiveness of the automatic rule recommendation. It is also demonstrated that by enforcing the recommended security rules through a camouflage engine, SARRE can effectively prevent information leakage and enable fine-grained protection over private data with very small performance overhead.",
"With the ubiquity of multi-core processors, software must make effective use of multiple cores to obtain good performance on modern hardware. One of the biggest roadblocks to this is load imbalance, or the uneven distribution of work across cores. We propose LIME, a framework for analyzing parallel programs and reporting the cause of load imbalance in application source code. This framework uses statistical techniques to pinpoint load imbalance problems stemming from both control flow issues (e.g., unequal iteration counts) and interactions between the application and hardware (e.g., unequal cache miss counts). We evaluate LIME on applications from widely used parallel benchmark suites, and show that LIME accurately reports the causes of load imbalance, their nature and origin in the code, and their relative importance.",
"Identifying vulnerabilities in software systems is crucial to minimizing the damages that result from malicious exploits and software failures. This often requires proper identification of vulnerable execution paths that contain program vulnerabilities or bugs. However, with rapid rise in software complexity, it has become notoriously difficult to identify such vulnerable paths through exhaustively searching the entire program execution space. In this paper, we propose StatSym, a novel, automated Statistics-Guided Symbolic Execution framework that integrates the swiftness of statistical inference and the rigorousness of symbolic execution techniques to achieve precision, agility and scalability in vulnerable program path discovery. Our solution first leverages statistical analysis of program runtime information to construct predicates that are indicative of potential vulnerability in programs. These statistically identified paths, along with the associated predicates, effectively drive a symbolic execution engine to verify the presence of vulnerable paths and reduce their time to solution. We evaluate StatSym on four real-world applications including polymorph, CTree, Grep and thttpd that come from diverse domains. Results show that StatSym is able to assist the symbolic executor, KLEE, in identifying the vulnerable paths for all of the four cases, whereas pure symbolic execution fails in three out of four applications due to memory space overrun.",
"",
"Different uses of memory protection schemes have different needs in terms of granularity. For example, heap security can benefit from chunk separation (by using protected \"padding\" boundaries) and meta-data protection. However, such protection can be done at different granularity (eg. per-word, per-block, or per-page), with different performance, cost and memory overhead tradeoffs for different applications. In this paper, we explore these tradeoffs for the purpose of heap security in order to discover whether the \"right\" granularity exists and how the granularity of protection affects design decisions.We evaluate such tradeoffs based on the current heap-security approaches in a single address spare operating system. The access control granularities we use are word, 8-byte, 16-byte, 32-byte, and page. We find that none of these schemes is optimal across all applications. In some applications, excessive padding degrades caching performance for coarse-granularity schemes, while in others, large-block permission changes introduce large overheads for finer granularities. To overcome these limitations, we propose a new two-granularity scheme, which uses word- and page-granularity protection to eliminate padding but allow fast page-size permission changes for large memory blocks. On all applications, this new scheme performs as well or better than the best single-granularity scheme. It also performs on par with the more complex Mondrian Memory Protection, which uses a complex trie structure and multiple permissions caching mechanisms to support a hierarchy of protection granularities.",
"In today's multicore processors, the last-level cache is often shared by multiple concurrently running processes to make efficient use of hardware resources. However, previous studies have shown that a shared cache is vulnerable to timing channel attacks that leak confidential information from one process to another. Static cache partitioning can eliminate the cache timing channels but incurs significant performance overhead. In this paper, we propose Secure Dynamic Cache Partitioning (SecDCP), a partitioning technique that defeats cache timing channel attacks. The SecDCP scheme changes the size of cache partitions at run time for better performance while preventing insecure information leakage between processes. For cache-sensitive multiprogram workloads, our experimental results show that SecDCP improves performance by up to 43 and by an average of 12.5 over static cache partitioning.",
"While multicore processors promise large performance benefits for parallel applications, writing these applications is notoriously difficult. Tuning a parallel application to achieve good performance, also known as performance debugging, is often more challenging than debugging the application for correctness. Parallel programs have many performance-related issues that are not seen in sequential programs. An increase in cache misses is one of the biggest challenges that programmers face. To minimize these misses, programmers must not only identify the source of the extra misses, but also perform the tricky task of determining if the misses are caused by interthread communication (i.e., coherence misses) and if so, whether they are caused by true or false sharing (since the solutions for these two are quite different). In this article, we propose a new programmer-centric definition of false sharing misses and describe our novel algorithm to perform coherence miss classification. We contrast our approach with existing data-centric definitions of false sharing. A straightforward implementation of our algorithm is too expensive to be incorporated in real hardware. Therefore, we explore the design space for low-cost hardware support that can classify coherence misses on-the-fly into true and false sharing misses, allowing existing performance counters and profiling tools to expose and attribute them. We find that our approximate schemes achieve good accuracy at only a fraction of the cost of the ideal scheme. Additionally, we demonstrate the usefulness of our work in a case study involving a real application."
]
} |
1902.04856 | 2913732222 | Typical person re-identification frameworks search for k best matches in a gallery of images that are often collected in varying conditions. The gallery may contain image sequences when re-identification is done on videos. However, such a process is time consuming as re-identification has to be carried out multiple times. In this paper, we extract spatio-temporal sequences of frames (referred to as tubes) of moving persons and apply a multi-stage processing to match a given query tube with a gallery of stored tubes recorded through other cameras. Initially, we apply a binary classifier to remove noisy images from the input query tube. In the next step, we use a key-pose detection-based query minimization. This reduces the length of the query tube by removing redundant frames. Finally, a 3-stage hierarchical re-identification framework is used to rank the output tubes as per the matching scores. Experiments with publicly available video re-identification datasets reveal that our framework is better than state-of-the-art methods. It ranks the tubes with an increased CMC accuracy of 6-8 across multiple datasets. Also, our method significantly reduces the number of false positives. A new video re-identification dataset, named Tube-based Reidentification Video Dataset (TRiViD), has been prepared with an aim to help the re-identification research community | Person re-identification applications are growing rapidly in numbers. However, humongeous growth in CCTV surveillance has thrown up various challenges to the re-identification research community. The primary challenges are to handle large volume of data @cite_38 @cite_0 , tracking in complex environment @cite_22 @cite_23 , presence of group @cite_5 , occlusion @cite_10 , varying pose and style across different cameras @cite_15 @cite_13 @cite_8 @cite_34 , etc. The process of Re-Id can be categorized as image-guided @cite_27 @cite_11 @cite_5 @cite_15 and video-guided @cite_36 @cite_35 @cite_29 @cite_32 @cite_17 . The image-guided methods typically use deep neural networks for feature representation and re-identification, whereas the video-guided methods typically use recurrent convolutional networks (RNN) to embed the temporal information such as optical flow @cite_29 , sequence of pose, etc. Table summarizes recent progress in person re-identification. In recent years, late fusion of different scores @cite_27 @cite_12 has shown significant improvement over the final ranking. Our method is similar to a typical delayed or late fusion guided method. We refine search results obtained using convolutional neural networks with the help of temporal correlation analysis. | {
"cite_N": [
"@cite_13",
"@cite_38",
"@cite_35",
"@cite_22",
"@cite_8",
"@cite_36",
"@cite_29",
"@cite_32",
"@cite_17",
"@cite_0",
"@cite_27",
"@cite_23",
"@cite_5",
"@cite_15",
"@cite_34",
"@cite_10",
"@cite_12",
"@cite_11"
],
"mid": [
"2798429327",
"2520433280",
"",
"2511791013",
"2769925982",
"2963852441",
"2779003141",
"2799185441",
"2781239370",
"2204750386",
"2776203024",
"",
"2798874329",
"",
"2963289251",
"2798590501",
"1920259731",
"2962706983"
],
"abstract": [
"Person re-identification (ReID) is an important task in the field of intelligent security. A key challenge is how to capture human pose variations, while existing benchmarks (i.e., Market1501, DukeMTMC-reID, CUHK03, etc.) do NOT provide sufficient pose coverage to train a robust ReID system. To address this issue, we propose a pose-transferrable person ReID framework which utilizes posetransferred sample augmentations (i.e., with ID supervision) to enhance ReID model training. On one hand, novel training samples with rich pose variations are generated via transferring pose instances from MARS dataset, and they are added into the target dataset to facilitate robust training. On the other hand, in addition to the conventional discriminator of GAN (i.e., to distinguish between REAL FAKE samples), we propose a novel guider sub-network which encourages the generated sample (i.e., with novel pose) towards better satisfying the ReID loss (i.e., cross-entropy ReID loss, triplet ReID loss). In the meantime, an alternative optimization procedure is proposed to train the proposed Generator-Guider-Discriminator network. Experimental results on Market-1501, DukeMTMC-reID and CUHK03 show that our method achieves great performance improvement, and outperforms most state-of-the-art methods without elaborate designing the ReID model.",
"This paper considers person re-identification (re-id) in videos. We introduce a new video re-id dataset, named Motion Analysis and Re-identification Set (MARS), a video extension of the Market-1501 dataset. To our knowledge, MARS is the largest video re-id dataset to date. Containing 1,261 IDs and around 20,000 tracklets, it provides rich visual information compared to image-based datasets. Meanwhile, MARS reaches a step closer to practice. The tracklets are automatically generated by the Deformable Part Model (DPM) as pedestrian detector and the GMMCP tracker. A number of false detection tracking results are also included as distractors which would exist predominantly in practical video databases. Extensive evaluation of the state-of-the-art methods including the space-time descriptors and CNN is presented. We show that CNN in classification mode can be trained from scratch using the consecutive bounding boxes of each identity. The learned CNN embedding outperforms other competing methods considerably and has good generalization ability on other video re-id datasets upon fine-tuning.",
"",
"To help accelerate progress in multi-target, multi-camera tracking systems, we present (i) a new pair of precision-recall measures of performance that treats errors of all types uniformly and emphasizes correct identification over sources of error; (ii) the largest fully-annotated and calibrated data set to date with more than 2 million frames of 1080 p, 60 fps video taken by 8 cameras observing more than 2,700 identities over 85 min; and (iii) a reference software system as a comparison baseline. We show that (i) our measures properly account for bottom-line identity match performance in the multi-camera setting; (ii) our data set poses realistic challenges to current trackers; and (iii) the performance of our system is comparable to the state of the art.",
"Person re identification is a challenging retrieval task that requires matching a person's acquired image across non overlapping camera views. In this paper we propose an effective approach that incorporates both the fine and coarse pose information of the person to learn a discriminative embedding. In contrast to the recent direction of explicitly modeling body parts or correcting for misalignment based on these, we show that a rather straightforward inclusion of acquired camera view and or the detected joint locations into a convolutional neural network helps to learn a very effective representation. To increase retrieval performance, re-ranking techniques based on computed distances have recently gained much attention. We propose a new unsupervised and automatic re-ranking framework that achieves state-of-the-art re-ranking performance. We show that in contrast to the current state-of-the-art re-ranking methods our approach does not require to compute new rank lists for each image pair (e.g., based on reciprocal neighbors) and performs well by using simple direct rank list based comparison or even by just using the already computed euclidean distances between the images. We show that both our learned representation and our re-ranking method achieve state-of-the-art performance on a number of challenging surveillance image and video datasets. The code is available online at: this https URL",
"Most of the proposed person re-identification algorithms conduct supervised training and testing on single labeled datasets with small size, so directly deploying these trained models to a large-scale real-world camera network may lead to poor performance due to underfitting. It is challenging to incrementally optimize the models by using the abundant unlabeled data collected from the target domain. To address this challenge, we propose an unsupervised incremental learning algorithm, TFusion, which is aided by the transfer learning of the pedestrians' spatio-temporal patterns in the target domain. Specifically, the algorithm firstly transfers the visual classifier trained from small labeled source dataset to the unlabeled target dataset so as to learn the pedestrians' spatial-temporal patterns. Secondly, a Bayesian fusion model is proposed to combine the learned spatio-temporal patterns with visual features to achieve a significantly improved classifier. Finally, we propose a learning-to-rank based mutual promotion procedure to incrementally optimize the classifiers based on the unlabeled data in the target domain. Comprehensive experiments based on multiple real surveillance datasets are conducted, and the results show that our algorithm gains significant improvement compared with the state-of-art cross-dataset unsupervised person re-identification algorithms.",
"Person re-identification is an important task in video surveillance systems. It can be formally defined as establishing the correspondence between images of a person taken from different cameras at different times. In this paper, we present a two stream convolutional neural network where each stream is a Siamese network. This architecture can learn spatial and temporal information separately. We also propose a weighted two stream training objective function which combines the Siamese cost of the spatial and temporal streams with the objective of predicting a person's identity. We evaluate our proposed method on the publicly available PRID2011 and iLIDS-VID datasets and demonstrate the efficacy of our proposed method. On average, the top rank matching accuracy is 4 higher than the accuracy achieved by the cross-view quadratic discriminant analysis used in combination with the hierarchical Gaussian descriptor (GOG+XQDA), and 5 higher than the recurrent neural network method.",
"We focus on the one-shot learning for video-based person re-Identification (re-ID). Unlabeled tracklets for the person re-ID tasks can be easily obtained by preprocessing, such as pedestrian detection and tracking. In this paper, we propose an approach to exploiting unlabeled tracklets by gradually but steadily improving the discriminative capability of the Convolutional Neural Network (CNN) feature representation via stepwise learning. We first initialize a CNN model using one labeled tracklet for each identity. Then we update the CNN model by the following two steps iteratively: 1. sample a few candidates with most reliable pseudo labels from unlabeled tracklets; 2. update the CNN model according to the selected data. Instead of the static sampling strategy applied in existing works, we propose a progressive sampling method to increase the number of the selected pseudo-labeled candidates step by step. We systematically investigate the way how we should select pseudo-labeled tracklets into the training set to make the best use of them. Notably, the rank-1 accuracy of our method outperforms the state-of-the-art method by 21.46 points (absolute, i.e., 62.67 vs. 41.21 ) on the MARS dataset, and 16.53 points on the DukeMTMC-VideoReID dataset1.",
"Multi-shot pedestrian re-identification problem is at the core of surveillance video analysis. It matches two tracks of pedestrians from different cameras. In contrary to existing works that aggregate single frames features by time series model such as recurrent neural network, in this paper, we propose an interpretable reinforcement learning based approach to this problem. Particularly, we train an agent to verify a pair of images at each time. The agent could choose to output the result (same or different) or request another pair of images to verify (unsure). By this way, our model implicitly learns the difficulty of image pairs, and postpone the decision when the model does not accumulate enough evidence. Moreover, by adjusting the reward for unsure action, we can easily trade off between speed and accuracy. In three open benchmarks, our method are competitive with the state-of-the-art methods while only using 3 to 6 images. These promising results demonstrate that our method is favorable in both efficiency and performance.",
"This paper contributes a new high quality dataset for person re-identification, named \"Market-1501\". Generally, current datasets: 1) are limited in scale, 2) consist of hand-drawn bboxes, which are unavailable under realistic settings, 3) have only one ground truth and one query image for each identity (close environment). To tackle these problems, the proposed Market-1501 dataset is featured in three aspects. First, it contains over 32,000 annotated bboxes, plus a distractor set of over 500K images, making it the largest person re-id dataset to date. Second, images in Market-1501 dataset are produced using the Deformable Part Model (DPM) as pedestrian detector. Third, our dataset is collected in an open system, where each identity has multiple images under each camera. As a minor contribution, inspired by recent advances in large-scale image search, this paper proposes an unsupervised Bag-of-Words descriptor. We view person re-identification as a special task of image search. In experiment, we show that the proposed descriptor yields competitive accuracy on VIPeR, CUHK03, and Market-1501 datasets, and is scalable on the large-scale 500k dataset.",
"Person re-identification is a challenge in video-based surveillance where the goal is to identify the same person in different camera views. In recent years, many algorithms have been proposed that approach this problem by designing suitable feature representations for images of persons or by training appropriate distance metrics that learn to distinguish between images of different persons. Aggregating the results from multiple algorithms for person re-identification is a relatively less-explored area of research. In this paper, we formulate an algorithm that maps the ranking process in a person re-identification algorithm to a problem in graph theory. We then extend this formulation to allow for the use of results from multiple algorithms to make a consensus-based decision for the person re-identification problem. The algorithm is unsupervised and takes into account only the matching scores generated by multiple algorithms for creating a consensus of results. Further, we show how the graph theoretic problem can be solved by a two-step process. First, we obtain a rough estimate of the solution using a greedy algorithm. Then, we extend the construction of the proposed graph so that the problem can be efficiently solved by means of Ant Colony Optimization, a heuristic path-searching algorithm for complex graphs. While we present the algorithm in the context of person reidentification, it can potentially be applied to the general problem of ranking items based on a consensus of multiple sets of scores or metric values.",
"",
"Person re-identification benefits greatly from deep neural networks (DNN) to learn accurate similarity metrics and robust feature embeddings. However, most of the current methods impose only local constraints for similarity learning. In this paper, we incorporate constraints on large image groups by combining the CRF with deep neural networks. The proposed method aims to learn the \"local similarity\" metrics for image pairs while taking into account the dependencies from all the images in a group, forming \"group similarities\". Our method involves multiple images to model the relationships among the local and global similarities in a unified CRF during training, while combines multi-scale local similarities as the predicted similarity in testing. We adopt an approximate inference scheme for estimating the group similarity, enabling end-to-end training. Extensive experiments demonstrate the effectiveness of our model that combines DNN and CRF for learning robust multi-scale local similarities. The overall results outperform those by state-of-the-arts with considerable margins on three widely-used benchmarks.",
"",
"Being a cross-camera retrieval task, person re-identification suffers from image style variations caused by different cameras. The art implicitly addresses this problem by learning a camera-invariant descriptor subspace. In this paper, we explicitly consider this challenge by introducing camera style (CamStyle) adaptation. CamStyle can serve as a data augmentation approach that smooths the camera style disparities. Specifically, with CycleGAN, labeled training images can be style-transferred to each camera, and, along with the original training samples, form the augmented training set. This method, while increasing data diversity against over-fitting, also incurs a considerable level of noise. In the effort to alleviate the impact of noise, the label smooth regularization (LSR) is adopted. The vanilla version of our method (without LSR) performs reasonably well on few-camera systems in which over-fitting often occurs. With LSR, we demonstrate consistent improvement in all systems regardless of the extent of over-fitting. We also report competitive accuracy compared with the state of the art. Code is available at: https: github.com zhunzhong07 CamStyle",
"Person re-identification (ReID) is the task of retrieving particular persons across different cameras. Despite its great progress in recent years, it is still confronted with challenges like pose variation, occlusion, and similar appearance among different persons. The large gap between training and testing performance with existing models implies the insufficiency of generalization. Considering this fact, we propose to augment the variation of training data by introducing Adversarially Occluded Samples. These special samples are both a) meaningful in that they resemble real-scene occlusions, and b) effective in that they are tough for the original model and thus provide the momentum to jump out of local optimum. We mine these samples based on a trained ReID model and with the help of network visualization techniques. Extensive experiments show that the proposed samples help the model discover new discriminative clues on the body and generalize much better at test time. Our strategy makes significant improvement over strong baselines on three large-scale ReID datasets, Market1501, CUHK03 and DukeMTMC-reID.",
"We propose an effective structured learning based approach to the problem of person re-identification which outperforms the current state-of-the-art on most benchmark data sets evaluated. Our framework is built on the basis of multiple low-level hand-crafted and high-level visual features. We then formulate two optimization algorithms, which directly optimize evaluation measures commonly used in person re-identification, also known as the Cumulative Matching Characteristic (CMC) curve. Our new approach is practical to many real-world surveillance applications as the re-identification performance can be concentrated in the range of most practical importance. The combination of these factors leads to a person re-identification system which outperforms most existing algorithms. More importantly, we advance state-of-the-art results on person re-identification by improving the rank-1 recognition rates from 40 to 50 on the iLIDS benchmark, 16 to 18 on the PRID2011 benchmark, 43 to 46 on the VIPeR benchmark, 34 to 53 on the CUHK01 benchmark and 21 to 62 on the CUHK03 benchmark.",
"Key to effective person re-identification (Re-ID) is modelling discriminative and view-invariant factors of person appearance at both high and low semantic levels. Recently developed deep Re-ID models either learn a holistic single semantic level feature representation and or require laborious human annotation of these factors as attributes. We propose Multi-Level Factorisation Net (MLFN), a novel network architecture that factorises the visual appearance of a person into latent discriminative factors at multiple semantic levels without manual annotation. MLFN is composed of multiple stacked blocks. Each block contains multiple factor modules to model latent factors at a specific level, and factor selection modules that dynamically select the factor modules to interpret the content of each input image. The outputs of the factor selection modules also provide a compact latent factor descriptor that is complementary to the conventional deeply learned features. MLFN achieves state-of-the-art results on three Re-ID datasets, as well as compelling results on the general object categorisation CIFAR-100 dataset."
]
} |
1902.04856 | 2913732222 | Typical person re-identification frameworks search for k best matches in a gallery of images that are often collected in varying conditions. The gallery may contain image sequences when re-identification is done on videos. However, such a process is time consuming as re-identification has to be carried out multiple times. In this paper, we extract spatio-temporal sequences of frames (referred to as tubes) of moving persons and apply a multi-stage processing to match a given query tube with a gallery of stored tubes recorded through other cameras. Initially, we apply a binary classifier to remove noisy images from the input query tube. In the next step, we use a key-pose detection-based query minimization. This reduces the length of the query tube by removing redundant frames. Finally, a 3-stage hierarchical re-identification framework is used to rank the output tubes as per the matching scores. Experiments with publicly available video re-identification datasets reveal that our framework is better than state-of-the-art methods. It ranks the tubes with an increased CMC accuracy of 6-8 across multiple datasets. Also, our method significantly reduces the number of false positives. A new video re-identification dataset, named Tube-based Reidentification Video Dataset (TRiViD), has been prepared with an aim to help the re-identification research community | @cite_17 & Sequential decision making has been used to identify each frame in a video Multi-shot Pedestrian Re-identification via Sequential Decision Making | {
"cite_N": [
"@cite_17"
],
"mid": [
"2781239370"
],
"abstract": [
"Multi-shot pedestrian re-identification problem is at the core of surveillance video analysis. It matches two tracks of pedestrians from different cameras. In contrary to existing works that aggregate single frames features by time series model such as recurrent neural network, in this paper, we propose an interpretable reinforcement learning based approach to this problem. Particularly, we train an agent to verify a pair of images at each time. The agent could choose to output the result (same or different) or request another pair of images to verify (unsure). By this way, our model implicitly learns the difficulty of image pairs, and postpone the decision when the model does not accumulate enough evidence. Moreover, by adjusting the reward for unsure action, we can easily trade off between speed and accuracy. In three open benchmarks, our method are competitive with the state-of-the-art methods while only using 3 to 6 images. These promising results demonstrate that our method is favorable in both efficiency and performance."
]
} |
1902.04694 | 2898941873 | From formal and practical analysis, we identify new challenges that self-adaptive systems pose to the process of quality assurance. When tackling these, the effort spent on various tasks in the process of software engineering is naturally re-distributed. We claim that all steps related to testing need to become self-adaptive to match the capabilities of the self-adaptive system-under-test. Otherwise, the adaptive system’s behavior might elude traditional variants of quality assurance. We thus propose the paradigm of scenario coevolution, which describes a pool of test cases and other constraints on system behavior that evolves in parallel to the (in part autonomous) development of behavior in the system-under-test. Scenario coevolution offers a simple structure for the organization of adaptive testing that allows for both human-controlled and autonomous intervention, supporting software engineering for adaptive systems on a procedural as well as technical level. | Many researchers and practitioners in recent years have already been concerned about the changes necessary to allow for solid and reliable software engineering processes for (self -)adaptive systems. Central challenges were collected in @cite_13 , where issues of quality assurance are already mentioned but the focus is more on bringing about complex adaptive behavior in the first place. The later research roadmap of @cite_16 puts a strong focus on interaction patterns of already adaptive systems (both between each other and with human developers) and already dedicates a section to verification and validation issues, being close in mind to the perspective of this work. We fall in line with the roadmap further specified in @cite_24 @cite_26 @cite_8 . | {
"cite_N": [
"@cite_26",
"@cite_8",
"@cite_24",
"@cite_16",
"@cite_13"
],
"mid": [
"2405769265",
"2626275188",
"2261430958",
"2142486130",
"1964137226"
],
"abstract": [
"We discuss key challenges of software engineering for distributed autonomous real-time systems and introduce a taxonomy for areas of interest with respect to the development of such systems.",
"Smart Cyber--Physical Systems (sCPS) are modern CPS systems that are engineered to seamlessly integrate a large number of computation and physical components; they need to control entities in their environment in a smart and collective way to achieve a high degree of effectiveness and efficiency. At the same time, these systems are supposed to be safe and secure, deal with environment dynamicity and uncertainty, cope with external threats, and optimize their behavior to achieve the best possible outcome. This \"smartness\" typically stems from highly cooperative behavior, self--awareness, self--adaptation, and selfoptimization. Most of the \"smartness\" is implemented in software, which makes the software one of the most complex and most critical constituents of sCPS. As the specifics of sCPS render traditional software engineering approaches not directly applicable, new and innovative approaches to software engineering of sCPS need to be sought. This paper reports on the results of the Second International Workshop on Software Engineering for Smart Cyber--Physical Systems (SEsCPS 2016), which specifically focuses on challenges and promising solutions in the area of software engineering for sCPS.",
"Cyber-Physical Systems (CPS) are large interconnected softwareintensive systems that influence, by sensing and actuating, the physical world. Examples are traffic management and power grids. One of the trends we observe is the need to endow such systems with the \"smart\" capabilities, typically in the form of selfawareness and self-adaptation, along with the traditional qualities of safety and dependability. These requirements combined with specifics of the domain of smart CPS -- such as large scale, the role of end-users, uncertainty, and open-endedness -- render traditional software engineering (SE) techniques not directly applicable; making systematic SE of smart CPS a challenging task. This paper reports on the results of the First International Workshop on Software Engineering of Smart Cyber-Physical Systems (SEsCPS 2015), where participants discussed characteristics, challenges and opportunities of SE for smart CPS, with the aim to outline an agenda for future research in this important area.",
"The goal of this roadmap paper is to summarize the state-of-the-art and identify research challenges when developing, deploying and managing self-adaptive software systems. Instead of dealing with a wide range of topics associated with the field, we focus on four essential topics of self-adaptation: design space for self-adaptive solutions, software engineering processes for self-adaptive systems, from centralized to decentralized control, and practical run-time verification & validation for self-adaptive systems. For each topic, we present an overview, suggest future directions, and focus on selected challenges. This paper complements and extends a previous roadmap on software engineering for self-adaptive systems published in 2009 covering a different set of topics, and reflecting in part on the previous paper. This roadmap is one of the many results of the Dagstuhl Seminar 10431 on Software Engineering for Self-Adaptive Systems, which took place in October 2010.",
"Software systems dealing with distributed applications in changing environments normally require human supervision to continue operation in all conditions. These (re-)configuring, troubleshooting, and in general maintenance tasks lead to costly and time-consuming procedures during the operating phase. These problems are primarily due to the open-loop structure often followed in software development. Therefore, there is a high demand for management complexity reduction, management automation, robustness, and achieving all of the desired quality requirements within a reasonable cost and time range during operation. Self-adaptive software is a response to these demands; it is a closed-loop system with a feedback loop aiming to adjust itself to changes during its operation. These changes may stem from the software system's self (internal causes, e.g., failure) or context (external events, e.g., increasing requests from users). Such a system is required to monitor itself and its context, detect significant changes, decide how to react, and act to execute such decisions. These processes depend on adaptation properties (called self-* properties), domain characteristics (context information or models), and preferences of stakeholders. Noting these requirements, it is widely believed that new models and frameworks are needed to design self-adaptive software. This survey article presents a taxonomy, based on concerns of adaptation, that is, how, what, when and where, towards providing a unified view of this emerging area. Moreover, as adaptive systems are encountered in many disciplines, it is imperative to learn from the theories and models developed in these other areas. This survey article presents a landscape of research in self-adaptive software by highlighting relevant disciplines and some prominent research projects. This landscape helps to identify the underlying research gaps and elaborates on the corresponding challenges."
]
} |
1902.04694 | 2898941873 | From formal and practical analysis, we identify new challenges that self-adaptive systems pose to the process of quality assurance. When tackling these, the effort spent on various tasks in the process of software engineering is naturally re-distributed. We claim that all steps related to testing need to become self-adaptive to match the capabilities of the self-adaptive system-under-test. Otherwise, the adaptive system’s behavior might elude traditional variants of quality assurance. We thus propose the paradigm of scenario coevolution, which describes a pool of test cases and other constraints on system behavior that evolves in parallel to the (in part autonomous) development of behavior in the system-under-test. Scenario coevolution offers a simple structure for the organization of adaptive testing that allows for both human-controlled and autonomous intervention, supporting software engineering for adaptive systems on a procedural as well as technical level. | While this work largely builds upon @cite_11 , there have been other approaches to formalize the notion of adaptivity: @cite_9 discusses high-level architectural patterns that form multiple inter-connected adaptation loops. In @cite_27 such feedback loops are based on the MAPE-K model @cite_12 . While these approaches largely focus on the formal construction of adaptive systems, there have also been approaches that assume a (more human-centric or at least tool-centric) software engineering perspective @cite_21 @cite_17 @cite_14 @cite_10 . We want to discuss two of those on greater detail: | {
"cite_N": [
"@cite_14",
"@cite_9",
"@cite_21",
"@cite_17",
"@cite_27",
"@cite_10",
"@cite_12",
"@cite_11"
],
"mid": [
"2521497315",
"2053316513",
"2055906872",
"1520900153",
"2025735304",
"2608365813",
"1990911977",
"107710200"
],
"abstract": [
"In order to accurately predict future states of a smart cyber-physical system, which can change its behavior to a large degree in response to environmental influences, the existence of precise models of the system and its surroundings is demandable. In machine engineering, ultra-high fidelity simulations have been developed to better understand both constraints in system design and possible consequences of external influences during the system's operation. These digital twins enable further applications in software design for complex cyber-physical systems as online planning methods can utilize good simulations to continuously optimize the system behavior, yielding a software architecture framework based on the information flow between the cyber-physical system, its physical environment and the digital twin model.",
"Self-adaptive software requires high dependability robustness, adaptability, and availability. The article describes an infrastructure supporting two simultaneous processes in self-adaptive software: system evolution, the consistent application of change over time, and system adaptation, the cycle of detecting changing circumstances and planning and deploying responsive modifications.",
"Self-adaptive software systems are capable of adjusting their behavior at run-time to achieve certain objectives. Such systems typically employ analytical models specified at design-time to assess their characteristics at run-time and make the appropriate adaptation decisions. However, prior to system's deployment, engineers often cannot foresee the changes in the environment, requirements, and system's operational profile. Therefore, any analytical model used in this setting relies on underlying assumptions that if not held at run-time make the analysis and hence the adaptation decisions inaccurate. We present and evaluate FeatUre-oriented Self-adaptatION (FUSION) framework, which aims to solve this problem by learning the impact of adaptation decisions on the system's goals. The framework (1) allows for automatic online fine-tuning of the adaptation logic to unanticipated conditions, (2) reduces the upfront effort required for building such systems, and (3) makes the run-time analysis of such systems very efficient.",
"In this paper, we discuss how for self-adaptive systems some activities that traditionally occur at development-time are moved to run-time. Responsibilities for these activities shift from software engineers to the system itself, causing the traditional boundary between development-time and run-time to blur. As a consequence, we argue how the traditional software engineering process needs to be reconceptualized to distinguish both development-time and run-time activities, and to support designers in taking decisions on how to properly engineer such systems.",
"The MAPE-K (Monitor-Analyze-Plan-Execute over a shared Knowledge) feedback loop is the most influential reference control model for autonomic and self-adaptive systems. This paper presents a conceptual and methodological framework for formal modeling, validating, and verifying distributed self-adaptive systems. We show how MAPE-K loops for self adaptation can be naturally specified in an abstract stateful language like Abstract State Machines. In particular, we exploit the concept of multi-agent Abstract State Machines to specify decentralized adaptation control by using MAPE computations. We support techniques for validating and verifying adaptation scenarios, and getting feedback of the correctness of the adaptation logic as implemented by the MAPE-K loops. In particular, a verification technique based on meta-properties is proposed to allow discovering unwanted interferences between MAPE-K loops at the early stages of the system design. As a proof-of concepts, we model and analyze a traffic monitoring system.",
"",
"A 2001 IBM manifesto observed that a looming software complexity crisis -caused by applications and environments that number into the tens of millions of lines of code - threatened to halt progress in computing. The manifesto noted the almost impossible difficulty of managing current and planned computing systems, which require integrating several heterogeneous environments into corporate-wide computing systems that extend into the Internet. Autonomic computing, perhaps the most attractive approach to solving this problem, creates systems that can manage themselves when given high-level objectives from administrators. Systems manage themselves according to an administrator's goals. New components integrate as effortlessly as a new cell establishes itself in the human body. These ideas are not science fiction, but elements of the grand challenge to create self-managing computing systems.",
"Ensembles--software-intensive systems with massive numbers of nodes or complex interactions between nodes, operating in open and non-deterministic environments and dynamically adapting to changes in their environment or requirements--pose many challenges to software development. We present first steps towards a system model for ensembles that allows us to express requirements using a wide variety of logics and fitness criteria over arbitrary preorders. Using this system model we then give a precise definition of \"black-box\" adaptation and show how this naturally leads to a preorder of adaptability on ensembles."
]
} |
1902.04694 | 2898941873 | From formal and practical analysis, we identify new challenges that self-adaptive systems pose to the process of quality assurance. When tackling these, the effort spent on various tasks in the process of software engineering is naturally re-distributed. We claim that all steps related to testing need to become self-adaptive to match the capabilities of the self-adaptive system-under-test. Otherwise, the adaptive system’s behavior might elude traditional variants of quality assurance. We thus propose the paradigm of scenario coevolution, which describes a pool of test cases and other constraints on system behavior that evolves in parallel to the (in part autonomous) development of behavior in the system-under-test. Scenario coevolution offers a simple structure for the organization of adaptive testing that allows for both human-controlled and autonomous intervention, supporting software engineering for adaptive systems on a procedural as well as technical level. | In the results of the (Autonomous Service Component ENSembles) project @cite_18 , the interplay between human developers and autonomous adaptation has been formalized in a life-cycle model featuring separate states for each the development progress of each respective feedback cycle. Classical software development tasks and self-adaptation (as well as self-monitoring and self-awareness) are regarded as equally powerful contributing mechanisms for the production of software. Both can be employed in junction to steer the development process. In addition, ASCENS built upon a (in parts) similar formal notion of adaptivity @cite_5 @cite_23 and sketched a connection between adaptivity in complex distributed systems and multi-goal multi-agent learning @cite_7 . | {
"cite_N": [
"@cite_5",
"@cite_18",
"@cite_7",
"@cite_23"
],
"mid": [
"1587853850",
"2344434315",
"1043938270",
"2135749407"
],
"abstract": [
"In this position paper we present a conceptual vision of adaptation, a key feature of autonomic systems. We put some stress on the role of control data and argue how some of the programming paradigms and models used for adaptive systems match with our conceptual framework.",
"The ASCENS project deals with designing systems as ensembles of adaptive components. Among the outputs of the ASCENS project are multiple tools that address particular issues in designing the ensembles, ranging from support for early stage formal modeling to runtime environment for executing and monitoring ensemble implementations. The goal of this chapter is to provide a compact description of the individual tools, which is supplemented by additional downloadable material on the project website.",
"Reasoning and learning for awareness and adaptation are challenging endeavors since cogitation has to be tightly integrated with action execution and reaction to unforeseen contingencies. After discussing the notion of awareness and presenting a classification scheme for awareness mechanisms, we introduce Extended Behavior Trees (XBTs), a novel modeling method for hierarchical, concurrent behaviors that allows the interleaving of reasoning, learning and actions. The semantics of XBTs are defined by a transformation to SCEL so that sophisticated synchronization strategies are straightforward to realize and different kinds of distributed, hierarchical learning and reasoning—from centrally coordinated to fully autonomic—can easily be expressed. We propose novel hierarchical reinforcement-learning strategies called Hierarchical (Lenient) Frequency-Adjusted Q-learning, that can be implemented using XBTs. Finally we discuss how XBTs can be used to define a multi-layer approach to learning, called teacher-student learning, that combines centralized and distributed learning in a seamless way.",
"The autonomic computing paradigm has been proposed to cope with size, complexity, and dynamism of contemporary software-intensive systems. The challenge for language designers is to devise appropriate abstractions and linguistic primitives to deal with the large dimension of systems and with their need to adapt to the changes of the working environment and to the evolving requirements. We propose a set of programming abstractions that permit us to represent behaviors, knowledge, and aggregations according to specific policies and to support programming context-awareness, self-awareness, and adaptation. Based on these abstractions, we define SCEL (Software Component Ensemble Language), a kernel language whose solid semantic foundations lay also the basis for formal reasoning on autonomic systems behavior. To show expressiveness and effectiveness of SCEL;’s design, we present a Java implementation of the proposed abstractions and show how it can be exploited for programming a robotics scenario that is used as a running example for describing the features and potential of our approach."
]
} |
1902.04694 | 2898941873 | From formal and practical analysis, we identify new challenges that self-adaptive systems pose to the process of quality assurance. When tackling these, the effort spent on various tasks in the process of software engineering is naturally re-distributed. We claim that all steps related to testing need to become self-adaptive to match the capabilities of the self-adaptive system-under-test. Otherwise, the adaptive system’s behavior might elude traditional variants of quality assurance. We thus propose the paradigm of scenario coevolution, which describes a pool of test cases and other constraints on system behavior that evolves in parallel to the (in part autonomous) development of behavior in the system-under-test. Scenario coevolution offers a simple structure for the organization of adaptive testing that allows for both human-controlled and autonomous intervention, supporting software engineering for adaptive systems on a procedural as well as technical level. | Besides the field of software engineering, the field of artificial intelligence research is currently (re-)discovering a lot of the same issues the discipline of of engineering for complex adaptive systems faced: The highly complex and opaque nature of machine learning algorithms and the resulting data structures often forces black-box testing and makes possible guarantees weak. When online learning is employed, the algorithm's behavior is subject to great variance and testing usually needs to work online as well. The seminal paper @cite_4 provides a good overview of the issues. When applying artificial intelligence to a large variety of products, rigorous engineering for this kind of software seems to be one of the major necessities lacking at the moment. | {
"cite_N": [
"@cite_4"
],
"mid": [
"2462906003"
],
"abstract": [
"Rapid progress in machine learning and artificial intelligence (AI) has brought increasing attention to the potential impacts of AI technologies on society. In this paper we discuss one such potential impact: the problem of accidents in machine learning systems, defined as unintended and harmful behavior that may emerge from poor design of real-world AI systems. We present a list of five practical research problems related to accident risk, categorized according to whether the problem originates from having the wrong objective function (\"avoiding side effects\" and \"avoiding reward hacking\"), an objective function that is too expensive to evaluate frequently (\"scalable supervision\"), or undesirable behavior during the learning process (\"safe exploration\" and \"distributional shift\"). We review previous work in these areas as well as suggesting research directions with a focus on relevance to cutting-edge AI systems. Finally, we consider the high-level question of how to think most productively about the safety of forward-looking applications of AI."
]
} |
1902.04521 | 2914828627 | In this paper we consider the task of detecting abnormal communication volume occurring at node-level in communication networks. The signal of the communication activity is modeled by means of a clique stream: each occurring communication event is instantaneous and activates an undirected subgraph spanning over a set of equally participating nodes. We present a probabilistic framework to model and assess the communication volume observed at any single node. Specifically, we employ non-parametric regression to learn the probability that a node takes part in a certain event knowing the set of other nodes that are involved. On the top of that, we present a concentration inequality around the estimated volume of events in which a node could participate, which in turn allows us to build an efficient and interpretable anomaly scoring function. Finally, the superior performance of the proposed approach is empirically demonstrated in real-world sensor network data, as well as using synthetic communication activity that is in accordance with that latter setting. | A feature-based approach for detecting anomalies in such time-series of graphs, is to compute several graph features for each @math , such as the node degree or centrality, and then apply standard anomaly detection techniques on the derived multivariate time-series of these features @cite_3 @cite_19 @cite_10 . More generally, the literature of anomaly detection in time-series of graphs varies in three aspects: : Semi-supervised (access to a dataset of system operation) @cite_25 @cite_23 @cite_1 or unsupervised (no label available) @cite_4 @cite_11 . : Probabilistic model-based @cite_6 @cite_26 @cite_30 @cite_7 @cite_9 @cite_27 @cite_22 , distance based @cite_4 , decomposition-based @cite_0 @cite_5 , compression-based @cite_13 , : Node Edge-level @cite_18 @cite_22 , subgraph-level @cite_27 , or whole network-level @cite_7 . The reader should refer to @cite_16 for a more detailed survey on anomaly detection in dynamic graphs. | {
"cite_N": [
"@cite_30",
"@cite_22",
"@cite_3",
"@cite_5",
"@cite_10",
"@cite_18",
"@cite_4",
"@cite_23",
"@cite_26",
"@cite_7",
"@cite_6",
"@cite_19",
"@cite_27",
"@cite_16",
"@cite_25",
"@cite_9",
"@cite_1",
"@cite_0",
"@cite_13",
"@cite_11"
],
"mid": [
"1896833666",
"2026565928",
"93424348",
"2119511382",
"2122646361",
"",
"",
"2109169614",
"2086600682",
"2724643438",
"2140273660",
"2026493302",
"2094990982",
"2093168265",
"2132870739",
"2163557584",
"2000976678",
"1572352401",
"2171604933",
""
],
"abstract": [
"Interactions among people or objects are often dynamic in nature and can be represented as a sequence of networks, each providing a snapshot of the interactions over a brief period of time. An important task in analyzing such evolving networks is change-point detection, in which we both identify the times at which the large-scale pattern of interactions changes fundamentally and quantify how large and what kind of change occurred. Here, we formalize for the first time the network change-point detection problem within an online probabilistic learning framework and introduce a method that can reliably solve it. This method combines a generalized hierarchical random graph model with a Bayesian hypothesis test to quantitatively determine if, when, and precisely how a change point has occurred. We analyze the detectability of our method using synthetic data with known change points of different types and magnitudes, and show that this method is more accurate than several previously used alternatives. Applied to two high-resolution evolving social networks, this method identifies a sequence of change points that align with known external \"shocks\" to these networks.",
"Learning the network structure of a large graph is computationally demanding, and dynamically monitoring the network over time for any changes in structure threatens to be more challenging still. This paper presents a two-stage method for anomaly detection in dynamic graphs: the first stage uses simple, conjugate Bayesian models for discrete time counting processes to track the pairwise links of all nodes in the graph to assess normality of behavior; the second stage applies standard network inference tools on a greatly reduced subset of potentially anomalous nodes. The utility of the method is demonstrated on simulated and real data sets.",
"Anomaly detection in multivariate time series is an important data mining task with applications to ecosystem modeling, network traffic monitoring, medical diagnosis, and other domains. This paper presents a robust algorithm for detecting anomalies in noisy multivariate time series data by employing a kernel matrix alignment method to capture the dependence relationships among variables in the time series. Anomalies are found by performing a random walk traversal on the graph induced by the aligned kernel matrix. We show that the algorithm is flexible enough to handle different types of time series anomalies including subsequence-based and local anomalies. Our framework can also be used to characterize the anomalies found in a target time series in terms of the anomalies present in other time series. We have performed extensive experiments to empirically demonstrate the effectiveness of our algorithm. A case study is also presented to illustrate the ability of the algorithm to detect ecosystem disturbances in Earth science data.",
"Modern applications such as Internet traffic, telecommunication records, and large-scale social networks generate massive amounts of data with multiple aspects and high dimensionalities. Tensors (i.e., multi-way arrays) provide a natural representation for such data. Consequently, tensor decompositions such as Tucker become important tools for summarization and analysis. One major challenge is how to deal with high-dimensional, sparse data. In other words, how do we compute decompositions of tensors where most of the entries of the tensor are zero. Specialized techniques are needed for computing the Tucker decompositions for sparse tensors because standard algorithms do not account for the sparsity of the data. As a result, a surprising phenomenon is observed by practitioners: Despite the fact that there is enough memory to store both the input tensors and the factorized output tensors, memory overflows occur during the tensor factorization process. To address this intermediate blowup problem, we propose Memory-Efficient Tucker (MET). Based on the available memory, MET adaptively selects the right execution strategy during the decomposition. We provide quantitative and qualitative evaluation of MET on real tensors. It achieves over 1000X space reduction without sacrificing speed; it also allows us to work with much larger tensors that were too big to handle before. Finally, we demonstrate a data mining case-study using MET.",
"Anomaly detection is an important problem that has been researched within diverse research areas and application domains. Many anomaly detection techniques have been specifically developed for certain application domains, while others are more generic. This survey tries to provide a structured and comprehensive overview of the research on anomaly detection. We have grouped existing techniques into different categories based on the underlying approach adopted by each technique. For each category we have identified key assumptions, which are used by the techniques to differentiate between normal and anomalous behavior. When applying a given technique to a particular domain, these assumptions can be used as guidelines to assess the effectiveness of the technique in that domain. For each category, we provide a basic anomaly detection technique, and then show how the different existing techniques in that category are variants of the basic technique. This template provides an easier and more succinct understanding of the techniques belonging to each category. Further, for each category, we identify the advantages and disadvantages of the techniques in that category. We also provide a discussion on the computational complexity of the techniques since it is an important issue in real application domains. We hope that this survey will provide a better understanding of the different directions in which research has been done on this topic, and how techniques developed in one area can be applied in domains for which they were not intended to begin with.",
"",
"",
"Given a probability measure P and a reference measure μ, one is often interested in the minimum μ-measure set with P-measure at least α. Minimum volume sets of this type summarize the regions of greatest probability mass of P, and are useful for detecting anomalies and constructing confidence regions. This paper addresses the problem of estimating minimum volume sets based on independent samples distributed according to P. Other than these samples, no other information is available regarding P, but the reference measure μ is assumed to be known. We introduce rules for estimating minimum volume sets that parallel the empirical risk minimization and structural risk minimization principles in classification. As in classification, we show that the performances of our estimators are controlled by the rate of uniform convergence of empirical to true probabilities over the class from which the estimator is drawn. Thus we obtain finite sample size performance bounds in terms of VC dimension and related quantities. We also demonstrate strong universal consistency, an oracle inequality, and rates of convergence. The proposed estimators are illustrated with histogram and decision tree set estimation rules.",
"The ability to detect change-points in a dynamic network or a time series of graphs is an increasingly important task in many applications of the emerging discipline of graph signal processing. This paper formulates change-point detection as a hypothesis testing problem in terms of a generative latent position model, focusing on the special case of the Stochastic Block Model time series. We analyze two classes of scan statistics, based on distinct underlying locality statistics presented in the literature. Our main contribution is the derivation of the limiting properties and power characteristics of the competing scan statistics. Performance is compared theoretically, on synthetic data, and empirically, on the Enron email corpus.",
"The increasing amount of data stored in the form of dynamic interactions between actors necessitates the use of methodologies to automatically extract relevant information. The interactions can be represented by dynamic networks in which most existing methods look for clusters of vertices to summarize the data. In this paper, a new framework is proposed in order to cluster the vertices while detecting change points in the intensities of the interactions. These change points are key in the understanding of the temporal interactions. The model used involves non-homogeneous Poisson point processes with cluster-dependent piecewise constant intensity functions and common discontinuity points. A variational expectation maximization algorithm is derived for inference. We show that the pruned exact linear time method, originally developed for change points detection in univariate time series, can be considered for the maximization step. This allows the detection of both the number of change points and their location. Experiments on artificial and real datasets are carried out, and the proposed approach is compared with related methods.",
"We introduce a theory of scan statistics on graphs and apply the ideas to the problem of anomaly detection in a time series of Enron email graphs.",
"In the statistics community, outlier detection for time series data has been studied for decades. Recently, with advances in hardware and software technology, there has been a large body of work on temporal outlier detection from a computational perspective within the computer science community. In particular, advances in hardware technology have enabled the availability of various forms of temporal data collection mechanisms, and advances in software technology have enabled a variety of data management mechanisms. This has fueled the growth of different kinds of data sets such as data streams, spatio-temporal data, distributed streams, temporal networks, and time series data, generated by a multitude of applications. There arises a need for an organized and detailed study of the work done in the area of outlier detection with respect to such temporal datasets. In this survey, we provide a comprehensive and structured overview of a large set of interesting outlier definitions for various forms of temporal data, novel techniques, and application scenarios in which specific definitions and techniques have been widely used.",
"We introduce a computationally scalable method for detecting small anomalous areas in a large, time-dependent computer network, motivated by the challenge of identifying intruders operating inside enterprise-sized computer networks. Time-series of communications between computers are used to detect anomalies, and are modeled using Markov models that capture the bursty, often human-caused behavior that dominates a large subset of the time-series. Anomalies in these time-series are common, and the network intrusions we seek involve coincident anomalies over multiple connected pairs of computers. We show empirically that each time-series is nearly always independent of the time-series of other pairs of communicating computers. This independence is used to build models of normal activity in local areas from the models of the individual time-series, and these local areas are designed to detect the types of intrusions we are interested in. We define a locality statistic calculated by testing for deviations from...",
"Anomaly detection is an important problem with multiple applications, and thus has been studied for decades in various research domains. In the past decade there has been a growing interest in anomaly detection in data represented as networks, or graphs, largely because of their robust expressiveness and their natural ability to represent complex relationships. Originally, techniques focused on anomaly detection in static graphs, which do not change and are capable of representing only a single snapshot of data. As real-world networks are constantly changing, there has been a shift in focus to dynamic graphs, which evolve over time.",
"Suppose you are given some data set drawn from an underlying probability distribution P and you want to estimate a \"simple\" subset S of input space such that the probability that a test point drawn from P lies outside of S equals some a priori specified value between 0 and 1. We propose a method to approach this problem by trying to estimate a function f that is positive on S and negative on the complement. The functional form of f is given by a kernel expansion in terms of a potentially small subset of the training data; it is regularized by controlling the length of the weight vector in an associated feature space. The expansion coefficients are found by solving a quadratic programming problem, which we do by carrying out sequential optimization over pairs of input patterns. We also provide a theoretical analysis of the statistical performance of our algorithm. The algorithm is a natural extension of the support vector algorithm to the case of unlabeled data.",
"A number of applications in social networks, telecommunications, and mobile computing create massive streams of graphs. In many such applications, it is useful to detect structural abnormalities which are different from the “typical” behavior of the underlying network. In this paper, we will provide first results on the problem of structural outlier detection in massive network streams. Such problems are inherently challenging, because the problem of outlier detection is specially challenging because of the high volume of the underlying network stream. The stream scenario also increases the computational challenges for the approach. We use a structural connectivity model in order to define outliers in graph streams. In order to handle the sparsity problem of massive networks, we dynamically partition the network in order to construct statistically robust models of the connectivity behavior. We design a reservoir sampling method in order to maintain structural summaries of the underlying network. These structural summaries are designed in order to create robust, dynamic and efficient models for outlier detection in graph streams. We present experimental results illustrating the effectiveness and efficiency of our approach.",
"A minimum volume (MV) set, at level @a, is a set G\"@a^* having minimum volume among all those sets containing at least @a probability mass. MV sets provide a natural notion of the 'central mass' of a distribution and, as such, have recently become popular as a tool for the detection of anomalies in multivariate data. Motivated by the fact that anomaly detection problems frequently arise in settings with temporally indexed measurements, we propose here a new method for the estimation of MV sets from dependent data. Our method is based on the concept of complexity-penalized estimation, extending recent work of Scott and Nowak for the case of independent and identically distributed measurements, and has both desirable theoretical properties and a practical implementation. Of particular note is the fact that, for a large class of stochastic processes, choice of an appropriate complexity penalty reduces to the selection of a single tuning parameter, which represents the data dependency of the underlying stochastic process. While in reality the dependence structure is unknown, we offer a data-dependent method for selecting this parameter, based on subsampling principles. Our work is motivated by and illustrated through an application to the detection of anomalous traffic levels in Internet traffic time series.",
"Given a large sparse graph, how can we find patterns and anomalies? Several important applications can be modeled as large sparse graphs, e.g., network traffic monitoring, research citation network analysis, social network analysis, and regulatory networks in genes. Low rank decompositions, such as SVD and CUR, are powerful techniques for revealing latent hidden variables and associated patterns from high dimensional data. However, those methods often ignore the sparsity property of the graph, and hence usually incur too high memory and computational cost to be practical. We propose a novel method, the Compact Matrix Decomposition (CMD), to compute sparse low rank approximations. CMD dramatically reduces both the computation cost and the space requirements over existing decomposition methods (SVD, CUR). Using CMD as the key building block, we further propose procedures to efficiently construct and analyze dynamic graphs from real-time application data. We provide theoretical guarantee for our methods, and present results on two real, large datasets, one on network flow data (100GB trace of 22K hosts over one month) and one on DBLP (200MB over 25 years). We show that CMD is often an order of magnitude more efficient than the state of the art (SVD and CUR): it is over 10X faster, but requires less than 1 10 of the space, for the same reconstruction accuracy. Finally, we demonstrate how CMD is used for detecting anomalies and monitoring timeevolving graphs, in which it successfully detects worm-like hierarchical scanning patterns in real network data.",
"We consider the problem of network anomaly detection in large distributed systems. In this setting, Principal Component Analysis (PCA) has been proposed as a method for discovering anomalies by continuously tracking the projection of the data onto a residual subspace. This method was shown to work well empirically in highly aggregated networks, that is, those with a limited number of large nodes and at coarse time scales. This approach, however, has scalability limitations. To overcome these limitations, we develop a PCA-based anomaly detector in which adaptive local data filters send to a coordinator just enough data to enable accurate global detection. Our method is based on a stochastic matrix perturbation analysis that characterizes the tradeoff between the accuracy of anomaly detection and the amount of data communicated over the network.",
""
]
} |
1902.04521 | 2914828627 | In this paper we consider the task of detecting abnormal communication volume occurring at node-level in communication networks. The signal of the communication activity is modeled by means of a clique stream: each occurring communication event is instantaneous and activates an undirected subgraph spanning over a set of equally participating nodes. We present a probabilistic framework to model and assess the communication volume observed at any single node. Specifically, we employ non-parametric regression to learn the probability that a node takes part in a certain event knowing the set of other nodes that are involved. On the top of that, we present a concentration inequality around the estimated volume of events in which a node could participate, which in turn allows us to build an efficient and interpretable anomaly scoring function. Finally, the superior performance of the proposed approach is empirically demonstrated in real-world sensor network data, as well as using synthetic communication activity that is in accordance with that latter setting. | In @cite_14 , the framework is presented for the representation of a dynamic graph as a stream where edges are being created and removed. Therein, the nodes are assumed to be fixed and the dynamics affect only the edges between them. An edge is characterized by a triplet @math noting two communicating nodes @math , @math , and a time interval @math which is not necessarily continuous (may even be a union of non-contiguous time intervals). In this work, we adopt this stream framework. In particular, as we will see, we represent the activity as a clique stream, and @math is always a finite union of singletons as edges appear instantaneously. | {
"cite_N": [
"@cite_14"
],
"mid": [
"2761051840"
],
"abstract": [
"Graph theory provides a language for studying the structure of relations, and it is often used to study interactions over time too. However, it poorly captures the both temporal and structural nature of interactions, that calls for a dedicated formalism. In this paper, we generalize graph concepts in order to cope with both aspects in a consistent way. We start with elementary concepts like density, clusters, or paths, and derive from them more advanced concepts like cliques, degrees, clustering coefficients, or connected components. We obtain a language to directly deal with interactions over time, similar to the language provided by graphs to deal with relations. This formalism is self-consistent: usual relations between different concepts are preserved. It is also consistent with graph theory: graph concepts are special cases of the ones we introduce. This makes it easy to generalize higher-level objects such as quotient graphs, line graphs, k-cores, and centralities. This paper also considers discrete versus continuous time assumptions, instantaneous links, and extensions to more complex cases."
]
} |
1902.04257 | 2911719076 | To widen their accessibility and increase their utility, intelligent agents must be able to learn complex behaviors as specified by (non-expert) human users. Moreover, they will need to learn these behaviors within a reasonable amount of time while efficiently leveraging the sparse feedback a human trainer is capable of providing. Recent work has shown that human feedback can be characterized as a critique of an agent's current behavior rather than as an alternative reward signal to be maximized, culminating in the COnvergent Actor-Critic by Humans (COACH) algorithm for making direct policy updates based on human feedback. Our work builds on COACH, moving to a setting where the agent's policy is represented by a deep neural network. We employ a series of modifications on top of the original COACH algorithm that are critical for successfully learning behaviors from high-dimensional observations, while also satisfying the constraint of obtaining reduced sample complexity. We demonstrate the effectiveness of our Deep COACH algorithm in the rich 3D world of Minecraft with an agent that learns to complete tasks by mapping from raw pixels to actions using only real-time human feedback in 10-15 minutes of interaction. | Lastly, we make a distinction between the HRL setting utilized in this work and the complementary area of learning from demonstration @cite_18 . Under the learning-from-demonstration paradigm, an agent is provided with a dataset of demonstrations (usually, as trajectories), which capture a desired behavior. Given such a dataset, a learner is meant to solve for the underlying policy (or associated reward function) that best aligns with the observed data. Great successes in this area include imitation learning @cite_10 , inverse reinforcement learning @cite_19 , and inverse optimal control @cite_25 . In scenarios where it is possible to obtain such valuable demonstration data, these kinds of learning-from-demonstration approaches could be substituted for the unsupervised pre-training phase of our Deep COACH algorithm, leading to a robust initial policy that a human trainer could gradually augment with live feedback as needed. | {
"cite_N": [
"@cite_19",
"@cite_18",
"@cite_10",
"@cite_25"
],
"mid": [
"1999874108",
"1986014385",
"1931877416",
"2290104316"
],
"abstract": [
"We consider learning in a Markov decision process where we are not explicitly given a reward function, but where instead we can observe an expert demonstrating the task that we want to learn to perform. This setting is useful in applications (such as the task of driving) where it may be difficult to write down an explicit reward function specifying exactly how different desiderata should be traded off. We think of the expert as trying to maximize a reward function that is expressible as a linear combination of known features, and give an algorithm for learning the task demonstrated by the expert. Our algorithm is based on using \"inverse reinforcement learning\" to try to recover the unknown reward function. We show that our algorithm terminates in a small number of iterations, and that even though we may never recover the expert's reward function, the policy output by the algorithm will attain performance close to that of the expert, where here performance is measured with respect to the expert's unknown reward function.",
"We present a comprehensive survey of robot Learning from Demonstration (LfD), a technique that develops policies from example state to action mappings. We introduce the LfD design choices in terms of demonstrator, problem space, policy derivation and performance, and contribute the foundations for a structure in which to categorize LfD research. Specifically, we analyze and categorize the multiple ways in which examples are gathered, ranging from teleoperation to imitation, as well as the various techniques for policy derivation, including matching functions, dynamics models and plans. To conclude we discuss LfD limitations and related promising areas for future research.",
"Sequential prediction problems such as imitation learning, where future observations depend on previous predictions (actions), violate the common i.i.d. assumptions made in statistical learning. This leads to poor performance in theory and often in practice. Some recent approaches provide stronger guarantees in this setting, but remain somewhat unsatisfactory as they train either non-stationary or stochastic policies and require a large number of iterations. In this paper, we propose a new iterative algorithm, which trains a stationary deterministic policy, that can be seen as a no regret algorithm in an online learning setting. We show that any such no regret algorithm, combined with additional reduction assumptions, must find a policy with good performance under the distribution of observations it induces in such sequential settings. We demonstrate that this new approach outperforms previous approaches on two challenging imitation learning problems and a benchmark sequence labeling problem.",
"Reinforcement learning can acquire complex behaviors from high-level specifications. However, defining a cost function that can be optimized effectively and encodes the correct task is challenging in practice. We explore how inverse optimal control (IOC) can be used to learn behaviors from demonstrations, with applications to torque control of high-dimensional robotic systems. Our method addresses two key challenges in inverse optimal control: first, the need for informative features and effective regularization to impose structure on the cost, and second, the difficulty of learning the cost function under unknown dynamics for high-dimensional continuous systems. To address the former challenge, we present an algorithm capable of learning arbitrary nonlinear cost functions, such as neural networks, without meticulous feature engineering. To address the latter challenge, we formulate an efficient sample-based approximation for MaxEnt IOC. We evaluate our method on a series of simulated tasks and real-world robotic manipulation problems, demonstrating substantial improvement over prior methods both in terms of task complexity and sample efficiency."
]
} |
1902.04446 | 2963611757 | Algorithms for computing All-Pairs Shortest-Paths (APSP) are critical building blocks underlying many practical applications. The standard sequential algorithms, such as Floyd-Warshall and Johnson, quickly become infeasible for large input graphs, necessitating parallel approaches. In this work, we provide detailed analysis of parallel APSP performance on distributed memory clusters with Apache Spark. The Spark model allows for a portable and easy to deploy distributed implementation, and hence is attractive from the end-user point of view. We propose four different APSP implementations for large undirected weighted graphs, which differ in complexity and degree of reliance on techniques outside of pure Spark API. We demonstrate that Spark is able to handle APSP problems with over 200,000 vertices on a 1024-core cluster, and can compete with a naive MPI-based solution. However, our best performing solver requires auxiliary shared persistent storage, and is over two times slower than optimized MPI-based solver. | @cite_11 , propose recursive 2D block-cyclic algorithm that achieves lower-bound on communication latency and bandwidth, which they next extend to 2.5D communication avoiding formulation. On machine with 24,576 cores, the algorithms maintain strong scaling for problems with @math nodes, and weak scaling for up to @math nodes. The paper is also noteworthy for its extensive review of distributed memory APSP solvers. The advantage of the recursive formulation is induced data locality, which directly contributes to improved performance @cite_11 . However, in Spark, the concept of data locality is much weaker, since Spark's runtime system has a significant freedom in scheduling where to materialize or move data for computations. Even though programmer has some control over the data placement (e.g., via partitioning functions), it does not have direct control over where computations will be executed. Therefore, in our work, we derive from the iterative formulation of the 2D block-cyclic algorithm, which can be found in @cite_24 , and is one of several methods tracing their origin to transitive closure @cite_8 . | {
"cite_N": [
"@cite_24",
"@cite_8",
"@cite_11"
],
"mid": [
"2035449679",
"1999399436",
"2107651134"
],
"abstract": [
"We propose a blocked version of Floyd's all-pairs shortest-paths algorithm. The blocked algorithm makes better utilization of cache than does Floyd's original algorithm. Experiments indicate that the blocked algorithm delivers a speedup (relative to the unblocked Floyd's algorithm) between 1.6 and 1.9 on a Sun Ultra Enterprise 4000 5000 for graphs that have between 480 and 3200 vertices. The measured speedup on an SGI O2 for graphs with between 240 and 1200 vertices is between 1.6 and 2.",
"",
"We consider distributed memory algorithms for the all-pairs shortest paths (APSP) problem. Scaling the APSP problem to high concurrencies requires both minimizing inter-processor communication as well as maximizing temporal data locality. The 2.5D APSP algorithm, which is based on the divide-and-conquer paradigm, satisfies both of these requirements: it can utilize any extra available memory to perform asymptotically less communication, and it is rich in semiring matrix multiplications, which have high temporal locality. We start by introducing a block-cyclic 2D (minimal memory) APSP algorithm. With a careful choice of block-size, this algorithm achieves known communication lower-bounds for latency and bandwidth. We extend this 2D block-cyclic algorithm to a 2.5D algorithm, which can use c extra copies of data to reduce the bandwidth cost by a factor of c1 2, compared to its 2D counterpart. However, the 2.5D algorithm increases the latency cost by c1 2. We provide a tighter lower bound on latency, which dictates that the latency overhead is necessary to reduce bandwidth along the critical path of execution. Our implementation achieves impressive performance and scaling to 24,576 cores of a Cray XE6 supercomputer by utilizing well-tuned intra-node kernels within the distributed memory algorithm."
]
} |
1902.04446 | 2963611757 | Algorithms for computing All-Pairs Shortest-Paths (APSP) are critical building blocks underlying many practical applications. The standard sequential algorithms, such as Floyd-Warshall and Johnson, quickly become infeasible for large input graphs, necessitating parallel approaches. In this work, we provide detailed analysis of parallel APSP performance on distributed memory clusters with Apache Spark. The Spark model allows for a portable and easy to deploy distributed implementation, and hence is attractive from the end-user point of view. We propose four different APSP implementations for large undirected weighted graphs, which differ in complexity and degree of reliance on techniques outside of pure Spark API. We demonstrate that Spark is able to handle APSP problems with over 200,000 vertices on a 1024-core cluster, and can compete with a naive MPI-based solution. However, our best performing solver requires auxiliary shared persistent storage, and is over two times slower than optimized MPI-based solver. | @cite_15 , give a solution of the transitive closure problem, and apply it to solving APSP on a GPU. They demonstrate 2-4 @math speedup over highly tuned CPU implementation, and report problems for up to @math . Another efficient GPU-based solution is given by in @cite_5 . However, the method applies only to planar graphs. Djidjev solves APSP on sparse graphs with up to one million nodes, using a combination of multi-core CPUs and two GPUs. While we could combine GPU acceleration with Spark, e.g., by using Python and Numba Just-in-Time (JIT) compiler @cite_20 , it is not clear how advantageous such approach would be (we provide more detailed discussion in Sections and ). | {
"cite_N": [
"@cite_5",
"@cite_15",
"@cite_20"
],
"mid": [
"2117758387",
"1980153429",
"2245493112"
],
"abstract": [
"We describe a new algorithm for solving the all-pairs shortest-path (APSP) problem for planar graphs and graphs with small separators that exploits the massive on-chip parallelism available in today's Graphics Processing Units (GPUs). Our algorithm, based on the Floyd-War shall algorithm, has near optimal complexity in terms of the total number of operations, while its matrix-based structure is regular enough to allow for efficient parallel implementation on the GPUs. By applying a divide-and-conquer approach, we are able to make use of multi-node GPU clusters, resulting in more than an order of magnitude speedup over the fastest known Dijkstra-based GPU implementation and a two-fold speedup over a parallel Dijkstra-based CPU implementation.",
"The all-pairs shortest-path problem is an intricate part in numerous practical applications. We describe a shared memory cache efficient GPU implementation to solve transitive closure and the all-pairs shortest-path problem on directed graphs for large datasets. The proposed algorithmic design utilizes the resources available on the NVIDIA G80 GPU architecture using the CUDA API. Our solution generalizes to handle graph sizes that are inherently larger then the DRAM memory available on the GPU. Experiments demonstrate that our method is able to significantly increase processing large graphs making our method applicable for bioinformatics, internet node traffic, social networking, and routing problems.",
"Dynamic, interpreted languages, like Python, are attractive for domain-experts and scientists experimenting with new ideas. However, the performance of the interpreter is often a barrier when scaling to larger data sets. This paper presents a just-in-time compiler for Python that focuses in scientific and array-oriented computing. Starting with the simple syntax of Python, Numba compiles a subset of the language into efficient machine code that is comparable in performance to a traditional compiled language. In addition, we share our experience in building a JIT compiler using LLVM[1]."
]
} |
1902.04446 | 2963611757 | Algorithms for computing All-Pairs Shortest-Paths (APSP) are critical building blocks underlying many practical applications. The standard sequential algorithms, such as Floyd-Warshall and Johnson, quickly become infeasible for large input graphs, necessitating parallel approaches. In this work, we provide detailed analysis of parallel APSP performance on distributed memory clusters with Apache Spark. The Spark model allows for a portable and easy to deploy distributed implementation, and hence is attractive from the end-user point of view. We propose four different APSP implementations for large undirected weighted graphs, which differ in complexity and degree of reliance on techniques outside of pure Spark API. We demonstrate that Spark is able to handle APSP problems with over 200,000 vertices on a 1024-core cluster, and can compete with a naive MPI-based solution. However, our best performing solver requires auxiliary shared persistent storage, and is over two times slower than optimized MPI-based solver. | Apache Spark frameworks, such as GraphX @cite_10 (no longer maintained) and GraphFrames @cite_18 , offer multi-source shortest-paths algorithms. These algorithms are simple extensions of the single source shortest paths solver in Pregel BSP model @cite_22 , and are not designed with APSP in mind. Our proposed implementations break away from the Pregel model because in the initial tests, GraphX was unable to handle any reasonable problem size, prompting us to investigate the solutions reported in this paper. | {
"cite_N": [
"@cite_18",
"@cite_10",
"@cite_22"
],
"mid": [
"2491983741",
"1982003698",
"2170616854"
],
"abstract": [
"Graph data is prevalent in many domains, but it has usually required specialized engines to analyze. This design is onerous for users and precludes optimization across complete workflows. We present GraphFrames, an integrated system that lets users combine graph algorithms, pattern matching and relational queries, and optimizes work across them. GraphFrames generalize the ideas in previous graph-on-RDBMS systems, such as GraphX and Vertexica, by letting the system materialize multiple views of the graph (not just the specific triplet views in these systems) and executing both iterative algorithms and pattern matching using joins. To make applications easy to write, GraphFrames provide a concise, declarative API based on the \"data frame\" concept in R that can be used for both interactive queries and standalone programs. Under this API, GraphFrames use a graph-aware join optimization algorithm across the whole computation that can select from the available views. We implement GraphFrames over Spark SQL, enabling parallel execution on Spark and integration with custom code. We find that GraphFrames make it easy to express end-to-end workflows and match or exceed the performance of standalone tools, while enabling optimizations across workflow steps that cannot occur in current systems. In addition, we show that GraphFrames' view abstraction makes it easy to further speed up interactive queries by registering the appropriate view, and that the combination of graph and relational data allows for other optimizations, such as attribute-aware partitioning.",
"From social networks to targeted advertising, big graphs capture the structure in data and are central to recent advances in machine learning and data mining. Unfortunately, directly applying existing data-parallel tools to graph computation tasks can be cumbersome and inefficient. The need for intuitive, scalable tools for graph computation has lead to the development of new graph-parallel systems (e.g., Pregel, PowerGraph) which are designed to efficiently execute graph algorithms. Unfortunately, these new graph-parallel systems do not address the challenges of graph construction and transformation which are often just as problematic as the subsequent computation. Furthermore, existing graph-parallel systems provide limited fault-tolerance and support for interactive data mining. We introduce GraphX, which combines the advantages of both data-parallel and graph-parallel systems by efficiently expressing graph computation within the Spark data-parallel framework. We leverage new ideas in distributed graph representation to efficiently distribute graphs as tabular data-structures. Similarly, we leverage advances in data-flow systems to exploit in-memory computation and fault-tolerance. We provide powerful new operations to simplify graph construction and transformation. Using these primitives we implement the PowerGraph and Pregel abstractions in less than 20 lines of code. Finally, by exploiting the Scala foundation of Spark, we enable users to interactively load, transform, and compute on massive graphs.",
"Many practical computing problems concern large graphs. Standard examples include the Web graph and various social networks. The scale of these graphs - in some cases billions of vertices, trillions of edges - poses challenges to their efficient processing. In this paper we present a computational model suitable for this task. Programs are expressed as a sequence of iterations, in each of which a vertex can receive messages sent in the previous iteration, send messages to other vertices, and modify its own state and that of its outgoing edges or mutate graph topology. This vertex-centric approach is flexible enough to express a broad set of algorithms. The model has been designed for efficient, scalable and fault-tolerant implementation on clusters of thousands of commodity computers, and its implied synchronicity makes reasoning about programs easier. Distribution-related details are hidden behind an abstract API. The result is a framework for processing large graphs that is expressive and easy to program."
]
} |
1902.04506 | 2911431914 | Within OSNs, many of our supposedly online friends may instead be fake accounts called social bots, part of large groups that purposely re-share targeted content. Here, we study retweeting behaviors on Twitter, with the ultimate goal of detecting retweeting social bots. We collect a dataset of 10M retweets. We design a novel visualization that we leverage to highlight benign and malicious patterns of retweeting activity. In this way, we uncover a 'normal' retweeting pattern that is peculiar of human-operated accounts, and 3 suspicious patterns related to bot activities. Then, we propose a bot detection technique that stems from the previous exploration of retweeting behaviors. Our technique, called Retweet-Buster (RTbust), leverages unsupervised feature extraction and clustering. An LSTM autoencoder converts the retweet time series into compact and informative latent feature vectors, which are then clustered with a hierarchical density-based algorithm. Accounts belonging to large clusters characterized by malicious retweeting patterns are labeled as bots. RTbust obtains excellent detection results, with F1 = 0.87, whereas competitors achieve F1 < 0.76. Finally, we apply RTbust to a large dataset of retweets, uncovering 2 previously unknown active botnets with hundreds of accounts. | Lastly, a trailblazing direction of research involves the application of adversarial machine learning to bot detection. Until now, bot detection has mostly followed a schema, where countermeasures are taken only after having witnessed evidence of bot mischiefs @cite_34 . Instead, in @cite_34 is proposed an adversarial framework enabling analyses. The framework has been later instantiated in @cite_13 by using evolutionary algorithms to test current detection techniques against a wider range of unseen, sophisticated bots. Another preliminary work in adversarial bot detection is described in @cite_17 . Among the positive outcomes of adversarial approaches to bot detection, is a more rapid understanding of the drawbacks of current detectors and the opportunity to gain insights into new features for achieving more robust and more reliable detectors. 0.04 0.04 | {
"cite_N": [
"@cite_34",
"@cite_13",
"@cite_17"
],
"mid": [
"2798428925",
"2900681641",
"2805358007"
],
"abstract": [
"",
"Abstract Since decades, genetic algorithms have been used as an effective heuristic to solve optimization problems. However, in order to be applied, genetic algorithms may require a string-based genetic encoding of information, which severely limited their applicability when dealing with online accounts. Remarkably, a behavioral modeling technique inspired by biological DNA has been recently proposed – and successfully applied – for monitoring and detecting spambots in Online Social Networks. In this so-called digital DNA representation, the behavioral lifetime of an account is encoded as a sequence of characters, namely a digital DNA sequence. In a previous work, the authors proposed to create synthetic digital DNA sequences that resemble the characteristics of the digital DNA sequences of real accounts. The combination of (i) the capability to model the accounts’ behaviors as digital DNA sequences, (ii) the possibility to create synthetic digital DNA sequences, and (iii) the evolutionary simulations allowed by genetic algorithms, open up the unprecedented opportunity to study – and even anticipate – the evolutionary patterns of modern social spambots. In this paper, we experiment with a novel ad-hoc genetic algorithm that allows to obtain behaviorally evolved spambots. By varying the different parameters of the genetic algorithm, we are able to evaluate the capability of the evolved spambots to escape a state-of-art behavior-based detection technique. Notably, despite such detection technique achieved excellent performances in the recent past, a number of our spambot evolutions manage to escape detection. Our analysis, if carried out at large-scale, would allow to proactively identify possible spambot evolutions capable of evading current detection techniques.",
"The identification of automated activitiy in social media, specifically the detection of social bots, has become one of the major tasks within the field of social media computation. Recently published classification algorithms and frameworks focus on the identification of single bot accounts. Within different Twitter experiments, we show that these classifiers can be bypassed by hybrid approaches, which on a first glance may motivate further research for more sophisticated techniques. However, we pose the question, whether the detection of single bot accounts is a necessary condition for identifying malicious, strategic attacks on public opinion. Or is it more productive to concentrate on detecting strategies?"
]
} |
1902.04259 | 2912937857 | Interactive Fiction (IF) games are complex textual decision making problems. This paper introduces NAIL, an autonomous agent for general parser-based IF games. NAIL won the 2018 Text Adventure AI Competition, where it was evaluated on twenty unseen games. This paper describes the architecture, development, and insights underpinning NAIL's performance. | There is also a growing body of work studying reinforcement learning agents in textual environments: LSTM-DQN introduced the idea of building a representation for the current state by processing narrative text with a recurrent network. Building on this idea, LSTM-DRQN added a second level recurrence over states, which was demonstrated to give the agent an ability to reason over locations visited in the past. In the realm of choice-based games, @cite_3 @cite_0 explored architectural choices for scoring different pre-defined actions. Another recent innovation included training a separate network to eliminate consideration of invalid actions . Finally, @cite_4 demonstrated end-to-end learning using an architecture that represents game state as a knowledge graph. | {
"cite_N": [
"@cite_0",
"@cite_4",
"@cite_3"
],
"mid": [
"2782698114",
"2902391430",
"2465628802"
],
"abstract": [
"The ability to learn optimal control policies in systems where action space is defined by sentences in natural language would allow many interesting real-world applications such as automatic optimisation of dialogue systems. Text-based games with multiple endings and rewards are a promising platform for this task, since their feedback allows us to employ reinforcement learning techniques to jointly learn text representations and control policies. We present a general text game playing agent, testing its generalisation and transfer learning performance and showing its ability to play multiple games at once. We also present pyfiction, an open-source library for universal access to different text games that could, together with our agent that implements its interface, serve as a baseline for future research.",
"",
"This paper introduces a novel architecture for reinforcement learning with deep neural networks designed to handle state and action spaces characterized by natural language, as found in text-based games. Termed a deep reinforcement relevance network (DRRN), the architecture represents action and state spaces with separate embedding vectors, which are combined with an interaction function to approximate the Q-function in reinforcement learning. We evaluate the DRRN on two popular text games, showing superior performance over other deep Q-learning architectures. Experiments with paraphrased action descriptions show that the model is extracting meaning rather than simply memorizing strings of text."
]
} |
1902.04248 | 2911334049 | We consider the two-group classification problem and propose a kernel classifier based on the optimal scoring framework. Unlike previous approaches, we provide theoretical guarantees on the expected risk consistency of the method. We also allow for feature selection by imposing structured sparsity using weighted kernels. We propose fully-automated methods for selection of all tuning parameters, and in particular adapt kernel shrinkage ideas for ridge parameter selection. Numerical studies demonstrate the superior classification performance of the proposed approach compared to existing nonparametric classifiers. | To our knowledge, the kernelized version of the optimal scoring problem has not been considered in the literature except for @cite_29 . Unlike @cite_29 , we fix the scores and provide theoretical guarantees for the method. Another major distinction of our method is the feature selection which is achieved by weighting the kernel and adding a sparsity penalty to the weights. | {
"cite_N": [
"@cite_29"
],
"mid": [
"2108146080"
],
"abstract": [
"Fishers linear discriminant analysis (LDA) is a classical multivariate technique both for dimension reduction and classification. The data vectors are transformed into a low dimensional subspace such that the class centroids are spread out as much as possible. In this subspace LDA works as a simple prototype classifier with linear decision boundaries. However, in many applications the linear boundaries do not adequately separate the classes. We present a nonlinear generalization of discriminant analysis that uses the kernel trick of representing dot products by kernel functions. The presented algorithm allows a simple formulation of the EM-algorithm in terms of kernel functions which leads to a unique concept for unsupervised mixture analysis, supervised discriminant analysis and semi-supervised discriminant analysis with partially unlabelled observations in feature spaces."
]
} |
1902.04147 | 2951877764 | Age-Related Macular Degeneration (AMD) is an asymptomatic retinal disease which may result in loss of vision. There is limited access to high-quality relevant retinal images and poor understanding of the features defining sub-classes of this disease. Motivated by recent advances in machine learning we specifically explore the potential of generative modeling, using Generative Adversarial Networks (GANs) and style transferring, to facilitate clinical diagnosis and disease understanding by feature extraction. We design an analytic pipeline which first generates synthetic retinal images from clinical images; a subsequent verification step is applied. In the synthesizing step we merge GANs (DCGANs and WGANs architectures) and style transferring for the image generation, whereas the verified step controls the accuracy of the generated images. We find that the generated images contain sufficient pathological details to facilitate ophthalmologists' task of disease classification and in discovery of disease relevant features. In particular, our system predicts the drusen and geographic atrophy sub-classes of AMD. Furthermore, the performance using CFP images for GANs outperforms the classification based on using only the original clinical dataset. Our results are evaluated using existing classifier of retinal diseases and class activated maps, supporting the predictive power of the synthetic images and their utility for feature extraction. Our code examples are available online. | Yet, a major issue has been the stability and convergence of training a GAN. Recent work @cite_19 demonstrated improved stability when using a Kantorovich-Rubinstein metric, which we have adopted in our training of the GAN for retinal images. Rapid advances have demonstrated that GANs generate realistic images, with a rich number of features. For example, GANs have been successfully applied for face generation @cite_5 , indoor scene reconstruction @cite_1 and person re-identification @cite_11 . Here we benefit from recent progress with GANs to generate new synthetic retinal disease images using both Deep Convolutional Generative Adversarial Networks (DCGANs) @cite_18 and Wasserstein GANs (WGANs) @cite_19 . These architectures utilize a convolutional decoder, and DCGAN enables the employment of large GANs using Convolutional Neural Networks (CNNs), resulting in stable training across various datasets. Finally, our use of WGANs improves the stability of learning,thereby avoiding known challenges such as model collapse @cite_19 . | {
"cite_N": [
"@cite_18",
"@cite_1",
"@cite_19",
"@cite_5",
"@cite_11"
],
"mid": [
"2173520492",
"2560722161",
"2739748921",
"2883861033",
"2774879330"
],
"abstract": [
"In recent years, supervised learning with convolutional networks (CNNs) has seen huge adoption in computer vision applications. Comparatively, unsupervised learning with CNNs has received less attention. In this work we hope to help bridge the gap between the success of CNNs for supervised learning and unsupervised learning. We introduce a class of CNNs called deep convolutional generative adversarial networks (DCGANs), that have certain architectural constraints, and demonstrate that they are a strong candidate for unsupervised learning. Training on various image datasets, we show convincing evidence that our deep convolutional adversarial pair learns a hierarchy of representations from object parts to scenes in both the generator and discriminator. Additionally, we use the learned features for novel tasks - demonstrating their applicability as general image representations.",
"Generation of 3D data by deep neural network has been attracting increasing attention in the research community. The majority of extant works resort to regular representations such as volumetric grids or collection of images, however, these representations obscure the natural invariance of 3D shapes under geometric transformations, and also suffer from a number of other issues. In this paper we address the problem of 3D reconstruction from a single image, generating a straight-forward form of output – point cloud coordinates. Along with this problem arises a unique and interesting issue, that the groundtruth shape for an input image may be ambiguous. Driven by this unorthordox output form and the inherent ambiguity in groundtruth, we design architecture, loss function and learning paradigm that are novel and effective. Our final solution is a conditional shape sampler, capable of predicting multiple plausible 3D point clouds from an input image. In experiments not only can our system outperform state-of-the-art methods on single image based 3D reconstruction benchmarks, but it also shows strong performance for 3D shape completion and promising ability in making multiple plausible predictions.",
"",
"Recent advances in Generative Adversarial Networks (GANs) have shown impressive results for task of facial expression synthesis. The most successful architecture is StarGAN, that conditions GANs’ generation process with images of a specific domain, namely a set of images of persons sharing the same expression. While effective, this approach can only generate a discrete number of expressions, determined by the content of the dataset. To address this limitation, in this paper, we introduce a novel GAN conditioning scheme based on Action Units (AU) annotations, which describes in a continuous manifold the anatomical facial movements defining a human expression. Our approach allows controlling the magnitude of activation of each AU and combine several of them. Additionally, we propose a fully unsupervised strategy to train the model, that only requires images annotated with their activated AUs, and exploit attention mechanisms that make our network robust to changing backgrounds and lighting conditions. Extensive evaluation show that our approach goes beyond competing conditional generators both in the capability to synthesize a much wider range of expressions ruled by anatomically feasible muscle movements, as in the capacity of dealing with images in the wild.",
"Person Re-identification (re-id) faces two major challenges: the lack of cross-view paired training data and learning discriminative identity-sensitive and view-invariant features in the presence of large pose variations. In this work, we address both problems by proposing a novel deep person image generation model for synthesizing realistic person images conditional on pose. The model is based on a generative adversarial network (GAN) and used specifically for pose normalization in re-id, thus termed pose-normalization GAN (PN-GAN). With the synthesized images, we can learn a new type of deep re-id feature free of the influence of pose variations. We show that this feature is strong on its own and highly complementary to features learned with the original images. Importantly, we now have a model that generalizes to any new re-id dataset without the need for collecting any training data for model fine-tuning, thus making a deep re-id model truly scalable. Extensive experiments on five benchmarks show that our model outperforms the state-of-the-art models, often significantly. In particular, the features learned on Market-1501 can achieve a Rank-1 accuracy of 68.67 on VIPeR without any model fine-tuning, beating almost all existing models fine-tuned on the dataset."
]
} |
1902.04147 | 2951877764 | Age-Related Macular Degeneration (AMD) is an asymptomatic retinal disease which may result in loss of vision. There is limited access to high-quality relevant retinal images and poor understanding of the features defining sub-classes of this disease. Motivated by recent advances in machine learning we specifically explore the potential of generative modeling, using Generative Adversarial Networks (GANs) and style transferring, to facilitate clinical diagnosis and disease understanding by feature extraction. We design an analytic pipeline which first generates synthetic retinal images from clinical images; a subsequent verification step is applied. In the synthesizing step we merge GANs (DCGANs and WGANs architectures) and style transferring for the image generation, whereas the verified step controls the accuracy of the generated images. We find that the generated images contain sufficient pathological details to facilitate ophthalmologists' task of disease classification and in discovery of disease relevant features. In particular, our system predicts the drusen and geographic atrophy sub-classes of AMD. Furthermore, the performance using CFP images for GANs outperforms the classification based on using only the original clinical dataset. Our results are evaluated using existing classifier of retinal diseases and class activated maps, supporting the predictive power of the synthetic images and their utility for feature extraction. Our code examples are available online. | Recently, researchers have used convolutional neural networks to generate images @cite_17 with different given style. The method makes use of a pre-trained network to optimize the image and its features. However, this method operates as a global optimization; therefore, generated image exhibit distortions and detailed parts cannot be presented on the transferred images. Meeting this challenge, recent work @cite_21 accomplished realistic image generation and style transferring. On the other hand, in @cite_18 , authors combined a convolutional neural network and GANs to generate new images, thus mitigating the impact of a limited number of features of pathological relevance in original images. While their work clearly improved the state-of-the-art methods, the technique may generate poor image samples or fails to converge. To ensure convergence and quality of generated images, we deploy a closed form solution for style transferring @cite_9 . | {
"cite_N": [
"@cite_9",
"@cite_18",
"@cite_21",
"@cite_17"
],
"mid": [
"2788095258",
"2173520492",
"2604721644",
"2475287302"
],
"abstract": [
"Photorealistic image style transfer algorithms aim at stylizing a content photo using the style of a reference photo with the constraint that the stylized photo should remains photorealistic. While several methods exist for this task, they tend to generate spatially inconsistent stylizations with noticeable artifacts. In addition, these methods are computationally expensive, requiring several minutes to stylize a VGA photo. In this paper, we present a novel algorithm to address the limitations. The proposed algorithm consists of a stylization step and a smoothing step. While the stylization step transfers the style of the reference photo to the content photo, the smoothing step encourages spatially consistent stylizations. Unlike existing algorithms that require iterative optimization, both steps in our algorithm have closed-form solutions. Experimental results show that the stylized photos generated by our algorithm are twice more preferred by human subjects in average. Moreover, our method runs 60 times faster than the state-of-the-art approach. Code and additional results are available at this https URL",
"In recent years, supervised learning with convolutional networks (CNNs) has seen huge adoption in computer vision applications. Comparatively, unsupervised learning with CNNs has received less attention. In this work we hope to help bridge the gap between the success of CNNs for supervised learning and unsupervised learning. We introduce a class of CNNs called deep convolutional generative adversarial networks (DCGANs), that have certain architectural constraints, and demonstrate that they are a strong candidate for unsupervised learning. Training on various image datasets, we show convincing evidence that our deep convolutional adversarial pair learns a hierarchy of representations from object parts to scenes in both the generator and discriminator. Additionally, we use the learned features for novel tasks - demonstrating their applicability as general image representations.",
"This paper introduces a deep-learning approach to photographic style transfer that handles a large variety of image content while faithfully transferring the reference style. Our approach builds upon the recent work on painterly transfer that separates style from the content of an image by considering different layers of a neural network. However, as is, this approach is not suitable for photorealistic style transfer. Even when both the input and reference images are photographs, the output still exhibits distortions reminiscent of a painting. Our contribution is to constrain the transformation from the input to the output to be locally affine in colorspace, and to express this constraint as a custom fully differentiable energy term. We show that this approach successfully suppresses distortion and yields satisfying photorealistic style transfers in a broad variety of scenarios, including transfer of the time of day, weather, season, and artistic edits.",
"Rendering the semantic content of an image in different styles is a difficult image processing task. Arguably, a major limiting factor for previous approaches has been the lack of image representations that explicitly represent semantic information and, thus, allow to separate image content from style. Here we use image representations derived from Convolutional Neural Networks optimised for object recognition, which make high level image information explicit. We introduce A Neural Algorithm of Artistic Style that can separate and recombine the image content and style of natural images. The algorithm allows us to produce new images of high perceptual quality that combine the content of an arbitrary photograph with the appearance of numerous wellknown artworks. Our results provide new insights into the deep image representations learned by Convolutional Neural Networks and demonstrate their potential for high level image synthesis and manipulation."
]
} |
1902.03588 | 2912860520 | Ad exchanges are widely used in platforms for online display advertising. Autonomous agents operating in these exchanges must learn policies for interacting profitably with a diverse, continually changing, but unknown market. We consider this problem from the perspective of a publisher, strategically interacting with an advertiser through a posted price mechanism. The learning problem for this agent is made difficult by the fact that information is censored, i.e., the publisher knows if an impression is sold but no other quantitative information. We address this problem using the Harsanyi-Bellman Ad Hoc Coordination (HBA) algorithm [1, 3], which conceptualises this interaction in terms of a Stochastic Bayesian Game and arrives at optimal actions by best responding with respect to probabilistic beliefs maintained over a candidate set of opponent behaviour profiles. We adapt and apply HBA to the censored information setting of ad exchanges. Also, addressing the case of stochastic opponents, we devise a strategy based on a Kaplan-Meier estimator for opponent modelling. We evaluate the proposed method using simulations wherein we show that HBA-KM achieves substantially better competitive ratio and lower variance of return than baselines, including a Q-learning agent and a UCB-based online learning agent, and comparable to the offline optimal algorithm. | Much has been written about ad display and sponsored search auctions, for each participating agent of the auction, either as publisher or advertiser. A key paper in the area of ad exchanges is that of Muthukrishnan @cite_10 , who laid out several research directions in this domain, informed by exposure to the practice in this domain. | {
"cite_N": [
"@cite_10"
],
"mid": [
"1554763265"
],
"abstract": [
"An emerging way to sell and buy display ads on the Internet is via ad exchanges. RightMedia [1], AdECN [2] and DoubleClick Ad Exchange [3] are examples of such real-time two-sided markets. We describe an abstraction of this market. Based on that abstraction, we present several research directions and discuss some insights."
]
} |
1902.03588 | 2912860520 | Ad exchanges are widely used in platforms for online display advertising. Autonomous agents operating in these exchanges must learn policies for interacting profitably with a diverse, continually changing, but unknown market. We consider this problem from the perspective of a publisher, strategically interacting with an advertiser through a posted price mechanism. The learning problem for this agent is made difficult by the fact that information is censored, i.e., the publisher knows if an impression is sold but no other quantitative information. We address this problem using the Harsanyi-Bellman Ad Hoc Coordination (HBA) algorithm [1, 3], which conceptualises this interaction in terms of a Stochastic Bayesian Game and arrives at optimal actions by best responding with respect to probabilistic beliefs maintained over a candidate set of opponent behaviour profiles. We adapt and apply HBA to the censored information setting of ad exchanges. Also, addressing the case of stochastic opponents, we devise a strategy based on a Kaplan-Meier estimator for opponent modelling. We evaluate the proposed method using simulations wherein we show that HBA-KM achieves substantially better competitive ratio and lower variance of return than baselines, including a Q-learning agent and a UCB-based online learning agent, and comparable to the offline optimal algorithm. | From the advertiser's perspective, @cite_15 study budget optimisation for sponsored search auctions. The authors cast the problem of budget optimisation as a Markov Decision Process (MDP) with censored observations and propose a learning algorithm based on Kaplan-Meier estimators. The authors also perform a large scale empirical demonstration on auction data from Microsoft, in order to demonstrate that their algorithm is extremely effective in practice. Another take on the advertiser's optimisation problem is the one by @cite_18 who study the design of a bidding agent in a marketplace for displaying ads. They provide algorithms and performance guarantees for both settings, while also experimentally evaluating their performance on a fitted log-normal distribution from data observed from the Right Media Exchange. | {
"cite_N": [
"@cite_15",
"@cite_18"
],
"mid": [
"2951545777",
"2158635658"
],
"abstract": [
"We consider the budget optimization problem faced by an advertiser participating in repeated sponsored search auctions, seeking to maximize the number of clicks attained under that budget. We cast the budget optimization problem as a Markov Decision Process (MDP) with censored observations, and propose a learning algorithm based on the wellknown Kaplan-Meier or product-limit estimator. We validate the performance of this algorithm by comparing it to several others on a large set of search auction data from Microsoft adCenter, demonstrating fast convergence to optimal performance.",
"Motivated by the emergence of auction-based marketplaces for display ads such as the Right Media Exchange, we study the design of a bidding agent that implements a display advertising campaign by bidding in such a marketplace. The bidding agent must acquire a given number of impressions with a given target spend, when the highest external bid in the marketplace is drawn from an unknown distribution P. The quantity and spend constraints arise from the fact that display ads are usually sold on a CPM basis. We consider both the full information setting, where the winning price in each auction is announced publicly, and the partially observable setting where only the winner obtains information about the distribution; these differ in the penalty incurred by the agent while attempting to learn the distribution. We provide algorithms for both settings, and prove performance guarantees using bounds on uniform closeness from statistics, and techniques from online learning. We experimentally evaluate these algorithms: both algorithms perform very well with respect to both target quantity and spend; further, our algorithm for the partially observable case performs nearly as well as that for the fully observable setting despite the higher penalty incurred during learning."
]
} |
1902.03588 | 2912860520 | Ad exchanges are widely used in platforms for online display advertising. Autonomous agents operating in these exchanges must learn policies for interacting profitably with a diverse, continually changing, but unknown market. We consider this problem from the perspective of a publisher, strategically interacting with an advertiser through a posted price mechanism. The learning problem for this agent is made difficult by the fact that information is censored, i.e., the publisher knows if an impression is sold but no other quantitative information. We address this problem using the Harsanyi-Bellman Ad Hoc Coordination (HBA) algorithm [1, 3], which conceptualises this interaction in terms of a Stochastic Bayesian Game and arrives at optimal actions by best responding with respect to probabilistic beliefs maintained over a candidate set of opponent behaviour profiles. We adapt and apply HBA to the censored information setting of ad exchanges. Also, addressing the case of stochastic opponents, we devise a strategy based on a Kaplan-Meier estimator for opponent modelling. We evaluate the proposed method using simulations wherein we show that HBA-KM achieves substantially better competitive ratio and lower variance of return than baselines, including a Q-learning agent and a UCB-based online learning agent, and comparable to the offline optimal algorithm. | Another literature that is closely relevant pertains to learning to interact in multiagent domains, with limited or no prior coordination. Related work includes @cite_7 @cite_6 @cite_13 . A key concept arising from this literature is that of modelling the opponent in terms of a hypothesis space of policies, that in a certain sense approximate the space from which that agent herself chooses the true policy. | {
"cite_N": [
"@cite_13",
"@cite_6",
"@cite_7"
],
"mid": [
"1606056663",
"2139993574",
"2951218345"
],
"abstract": [
"As autonomous agents proliferate in the real world, both in software and robotic settings, they will increasingly need to band together for cooperative activities with previously unfamiliar teammates. In such ad hoc team settings, team strategies cannot be developed a priori. Rather, an agent must be prepared to cooperate with many types of teammates: it must collaborate without pre-coordination. This paper challenges the AI community to develop theory and to implement prototypes of ad hoc team agents. It defines the concept of ad hoc team agents, specifies an evaluation paradigm, and provides examples of possible theoretical and empirical approaches to challenge. The goal is to encourage progress towards this ambitious, newly realistic, and increasingly important research goal.",
"The concept of creating autonomous agents capable of exhibiting ad hoc teamwork was recently introduced as a challenge to the AI, and specifically to the multiagent systems community. An agent capable of ad hoc teamwork is one that can effectively cooperate with multiple potential teammates on a set of collaborative tasks. Previous research has investigated theoretically optimal ad hoc teamwork strategies in restrictive settings. This paper presents the first empirical study of ad hoc teamwork in a more open, complex teamwork domain. Specifically, we evaluate a range of effective algorithms for on-line behavior generation on the part of a single ad hoc team agent that must collaborate with a range of possible teammates in the pursuit domain.",
"The ad hoc coordination problem is to design an autonomous agent which is able to achieve optimal flexibility and efficiency in a multiagent system with no mechanisms for prior coordination. We conceptualise this problem formally using a game-theoretic model, called the stochastic Bayesian game, in which the behaviour of a player is determined by its private information, or type. Based on this model, we derive a solution, called Harsanyi-Bellman Ad Hoc Coordination (HBA), which utilises the concept of Bayesian Nash equilibrium in a planning procedure to find optimal actions in the sense of Bellman optimal control. We evaluate HBA in a multiagent logistics domain called level-based foraging, showing that it achieves higher flexibility and efficiency than several alternative algorithms. We also report on a human-machine experiment at a public science exhibition in which the human participants played repeated Prisoner's Dilemma and Rock-Paper-Scissors against HBA and alternative algorithms, showing that HBA achieves equal efficiency and a significantly higher welfare and winning rate."
]
} |
1902.03501 | 2930304691 | The increasing adoption of machine learning tools has led to calls for accountability via model interpretability. But what does it mean for a machine learning model to be interpretable by humans, and how can this be assessed? We focus on two definitions of interpretability that have been introduced in the machine learning literature: simulatability (a user's ability to run a model on a given input) and "what if" local explainability (a user's ability to correctly indicate the outcome to a model under local changes to the input). Through a user study with 1000 participants, we test whether humans perform well on tasks that mimic the definitions of simulatability and "what if" local explainability on models that are typically considered locally interpretable. We find evidence consistent with the common intuition that decision trees and logistic regression models are interpretable and are more interpretable than neural networks. We propose a metric - the runtime operation count on the simulatability task - to indicate the relative interpretability of models and show that as the number of operations increases the users' accuracy on the local interpretability tasks decreases. | Although we are unaware of existing metrics for the local interpretability of a general model, many measures developed by the program analysis community aim at assessing the understandability of a general program, which could be seen as metrics for global interpretability. For example, the counts the number of independent paths through a program using its control flow graph @cite_4 . Despite calls for more experimentally grounded assessments of interpretability @cite_1 @cite_11 , there have so far been few user studies focusing on the interpretability of machine learning models. Perhaps most similar to this paper, Poursabzi- @cite_6 measure how the number of features and model transparency affect trust, simulatability, and mistake detection using randomized user studies on a similar scale to what we will consider here. While we focus on assessing model types, Poursabzi- vary across general model attributes (e.g. black box vs. clear). They find that clear models, meaning the inner calculations of the models are displayed to the user, are best simulated. They also determined that the model type did not affect the trust the user had in the model nor improved users' ability to correct their mistakes. | {
"cite_N": [
"@cite_1",
"@cite_4",
"@cite_6",
"@cite_11"
],
"mid": [
"2594475271",
"",
"2788362053",
"2795530988"
],
"abstract": [
"As machine learning systems become ubiquitous, there has been a surge of interest in interpretable machine learning: systems that provide explanation for their outputs. These explanations are often used to qualitatively assess other criteria such as safety or non-discrimination. However, despite the interest in interpretability, there is very little consensus on what interpretable machine learning is and how it should be measured. In this position paper, we first define interpretability and describe when interpretability is needed (and when it is not). Next, we suggest a taxonomy for rigorous evaluation and expose open questions towards a more rigorous science of interpretable machine learning.",
"",
"Despite a growing body of research focused on creating interpretable machine learning methods, there have been few empirical studies verifying whether interpretable methods achieve their intended effects on end users. We present a framework for assessing the effects of model interpretability on users via pre-registered experiments in which participants are shown functionally identical models that vary in factors thought to influence interpretability. Using this framework, we ran a sequence of large-scale randomized experiments, varying two putative drivers of interpretability: the number of features and the model transparency (clear or black-box). We measured how these factors impact trust in model predictions, the ability to simulate a model, and the ability to detect a model's mistakes. We found that participants who were shown a clear model with a small number of features were better able to simulate the model's predictions. However, we found no difference in multiple measures of trust and found that clear models did not improve the ability to correct mistakes. These findings suggest that interpretability research could benefit from more emphasis on empirically verifying that interpretable models achieve all their intended effects.",
"Advances in artificial intelligence, sensors and big data management have far-reaching societ al impacts. As these systems augment our everyday lives, it becomes increasing-ly important for people to understand them and remain in control. We investigate how HCI researchers can help to develop accountable systems by performing a literature analysis of 289 core papers on explanations and explaina-ble systems, as well as 12,412 citing papers. Using topic modeling, co-occurrence and network analysis, we mapped the research space from diverse domains, such as algorith-mic accountability, interpretable machine learning, context-awareness, cognitive psychology, and software learnability. We reveal fading and burgeoning trends in explainable systems, and identify domains that are closely connected or mostly isolated. The time is ripe for the HCI community to ensure that the powerful new autonomous systems have intelligible interfaces built-in. From our results, we propose several implications and directions for future research to-wards this goal."
]
} |
1902.03830 | 2926762703 | Intrinsic Image Decomposition (IID) is a challenging and interesting computer vision problem with various applications in several fields. We present novel semantic priors and an integrated approach for single image IID that involves analyzing image at three hierarchical context levels. Local context priors capture scene properties at each pixel within a small neighbourhood. Mid-level context priors encode object level semantics. Global context priors establish correspondences at the scene level. Our semantic priors are designed on both fixed and flexible regions, using selective search method and Convolutional Neural Network features. Our IID method is an iterative multistage optimization scheme and consists of two complementary formulations: @math smoothing for shading and @math sparsity for reflectance. Experiments and analysis of our method indicate the utility of our semantic priors and structured hierarchical analysis in an IID framework. We compare our method with other contemporary IID solutions and show results with lesser artifacts. Finally, we highlight that proper choice and encoding of prior knowledge can produce competitive results even when compared to end-to-end deep learning IID methods, signifying the importance of such priors. We believe that the insights and techniques presented in this paper would be useful in the future IID research. | Several IID methods depend on auxiliary scene data in various forms. @cite_18 , @cite_6 and @cite_4 take an RGBD image input and use depth to establish structural correspondences between image pixels. @cite_25 require user annotations in the form of scribbles marking constant reflectance regions, as auxiliary information. For videos, @cite_7 use optical flow and enforce temporal reflectance consistency constraint between frames. Similarly @cite_12 use multiple views and enforce spatial reflectance consistency by identifying corresponding scene points across images. This idea is also employed by @cite_22 for reflectance consistency between multiple illumination images. @cite_16 base their method on the similar idea and use focal stacks as auxiliary information by substituting it for depth. @cite_3 and @cite_13 use diverse photo collections to establish correspondences between image regions to build constraints. The common idea behind these methods is to approximate textural and shape similarities using the auxiliary information. The necessity to acquire additional input data is a major drawback of such methods. | {
"cite_N": [
"@cite_18",
"@cite_4",
"@cite_22",
"@cite_7",
"@cite_6",
"@cite_3",
"@cite_16",
"@cite_13",
"@cite_25",
"@cite_12"
],
"mid": [
"2581345",
"2117751343",
"2136748901",
"",
"2101856619",
"2020733757",
"2566721894",
"2146721395",
"2057552653",
"2091247382"
],
"abstract": [
"While intrinsic image decomposition has been studied extensively during the past a few decades, it is still a challenging problem. This is partly because commonly used constraints on shading and reflectance are often too restrictive to capture an important property of natural images, i.e., rich textures. In this paper, we propose a novel image model for handling textures in intrinsic image decomposition, which enables us to produce high quality results even with simple constraints. We also propose a novel constraint based on surface normals obtained from an RGB-D image. Assuming Lambertian surfaces, we formulate the constraint based on a locally linear embedding framework to promote local and global consistency on the shading layer. We demonstrate that combining the novel texture-aware image model and the novel surface normal based constraint can produce superior results to existing approaches.",
"In this paper we extend the “shape, illumination and reflectance from shading” (SIRFS) model [3, 4], which recovers intrinsic scene properties from a single image. Though SIRFS performs well on images of segmented objects, it performs poorly on images of natural scenes, which contain occlusion and spatially-varying illumination. We therefore present Scene-SIRFS, a generalization of SIRFS in which we have a mixture of shapes and a mixture of illuminations, and those mixture components are embedded in a “soft” segmentation of the input image. We additionally use the noisy depth maps provided by RGB-D sensors (in this case, the Kinect) to improve shape estimation. Our model takes as input a single RGB-D image and produces as output an improved depth map, a set of surface normals, a reflectance image, a shading image, and a spatially varying model of illumination. The output of our model can be used for graphics applications, or for any application involving RGB-D images.",
"Intrinsic images are a useful midlevel description of scenes proposed by H.G. Barrow and J.M. Tenenbaum (1978). An image is de-composed into two images: a reflectance image and an illumination image. Finding such a decomposition remains a difficult problem in computer vision. We focus on a slightly, easier problem: given a sequence of T images where the reflectance is constant and the illumination changes, can we recover T illumination images and a single reflectance image? We show that this problem is still imposed and suggest approaching it as a maximum-likelihood estimation problem. Following recent work on the statistics of natural images, we use a prior that assumes that illumination images will give rise to sparse filter outputs. We show that this leads to a simple, novel algorithm for recovering reflectance images. We illustrate the algorithm's performance on real and synthetic image sequences.",
"",
"We present a model for intrinsic decomposition of RGB-D images. Our approach analyzes a single RGB-D image and estimates albedo and shading fields that explain the input. To disambiguate the problem, our model estimates a number of components that jointly account for the reconstructed shading. By decomposing the shading field, we can build in assumptions about image formation that help distinguish reflectance variation from shading. These assumptions are expressed as simple nonlocal regularizers. We evaluate the model on real-world images and on a challenging synthetic dataset. The experimental results demonstrate that the presented approach outperforms prior models for intrinsic decomposition of RGB-D images.",
"An intrinsic image is a decomposition of a photo into an illumination layer and a reflectance layer, which enables powerful editing such as the alteration of an object's material independently of its illumination. However, decomposing a single photo is highly under-constrained and existing methods require user assistance or handle only simple scenes. In this paper, we compute intrinsic decompositions using several images of the same scene under different viewpoints and lighting conditions. We use multi-view stereo to automatically reconstruct 3D points and normals from which we derive relationships between reflectance values at different locations, across multiple views and consequently different lighting conditions. We use robust estimation to reliably identify reflectance ratios between pairs of points. From these, we infer constraints for our optimization and enforce a coherent solution across multiple views and illuminations. Our results demonstrate that this constrained optimization yields high-quality and coherent intrinsic decompositions of complex scenes. We illustrate how these decompositions can be used for image-based illumination transfer and transitions between views with consistent lighting.",
"In this paper, we presents a novel method (RGBF-IID) for intrinsic image decomposition of a wild scene without any restrictions on the complexity, illumination or scale of the image. We use focal stacks of the scene as input. A focal stack captures a scene at varying focal distances. Since focus depends on distance to the object, this representation has information beyond an RGB image towards an RGBD image with depth. We call our representation an RGBF image to highlight this. We use a robust focus measure and generalized random walk algorithm to compute dense probability maps across the stack. These maps are used to define sparse local and global pixel neighbourhoods, adhering to the structure of the underlying 3D scene. We use these neighbourhood correspondences with standard chromaticity assumptions as constraints in an optimization system. We present our results on both indoor and outdoor scenes using manually captured stacks of random objects under natural as well as artificial lighting conditions. We also test our system on a larger dataset of synthetically generated focal stacks from NYUv2 and MPI Sintel datasets and show competitive performance against current state-of-the-art IID methods that use RGBD images. Our method provides a strong evidence for the potential of RGBF modality in place of RGBD in computer vision.",
"Separating a photograph into its reflectance and illumination intrinsic images is a fundamentally ambiguous problem, and state-of-the-art algorithms combine sophisticated reflectance and illumination priors with user annotations to create plausible results. However, these algorithms cannot be easily extended to videos for two reasons: first, naively applying algorithms designed for single images to videos produce results that are temporally incoherent; second, effectively specifying user annotations for a video requires interactive feedback, and current approaches are orders of magnitudes too slow to support this. We introduce a fast and temporally consistent algorithm to decompose video sequences into their reflectance and illumination components. Our algorithm uses a hybrid e2ep formulation that separates image gradients into smooth illumination and sparse reflectance gradients using look-up tables. We use a multi-scale parallelized solver to reconstruct the reflectance and illumination from these gradients while enforcing spatial and temporal reflectance constraints and user annotations. We demonstrate that our algorithm automatically produces reasonable results, that can be interactively refined by users, at rates that are two orders of magnitude faster than existing tools, to produce high-quality decompositions for challenging real-world video sequences. We also show how these decompositions can be used for a number of video editing applications including recoloring, retexturing, illumination editing, and lighting-aware compositing.",
"We introduce a method to compute intrinsic images for a multiview set of outdoor photos with cast shadows, taken under the same lighting. We use an automatic 3D reconstruction from these photos and the sun direction as input and decompose each image into reflectance and shading layers, despite the inaccuracies and missing data of the 3D model. Our approach is based on two key ideas. First, we progressively improve the accuracy of the parameters of our image formation model by performing iterative estimation and combining 3D lighting simulation with 2D image optimization methods. Second, we use the image formation model to express reflectance as a function of discrete visibility values for shadow and light, which allows to introduce a robust visibility classifier for pairs of points in a scene. This classifier is used for shadow labeling, allowing to compute high-quality reflectance and shading layers. Our multiview intrinsic decomposition is of sufficient quality to allow relighting of the input images. We create shadow-caster geometry which preserves shadow silhouettes and, using the intrinsic layers, we can perform multiview relighting with moving cast shadows. We present results on several multiview datasets, and show how it is now possible to perform image-based rendering with changing illumination conditions.",
"Intrinsic images aim at separating an image into its reflectance and illumination components to facilitate further analysis or manipulation. This separation is severely ill posed and the most successful methods rely on user indications or precise geometry to resolve the ambiguities inherent to this problem. In this paper, we propose a method to estimate intrinsic images from multiple views of an outdoor scene without the need for precise geometry and with a few manual steps to calibrate the input. We use multiview stereo to automatically reconstruct a 3D point cloud of the scene. Although this point cloud is sparse and incomplete, we show that it provides the necessary information to compute plausible sky and indirect illumination at each 3D point. We then introduce an optimization method to estimate sun visibility over the point cloud. This algorithm compensates for the lack of accurate geometry and allows the extraction of precise shadows in the final image. We finally propagate the information computed over the sparse point cloud to every pixel in the photograph using image-guided propagation. Our propagation not only separates reflectance from illumination, but also decomposes the illumination into a sun, sky, and indirect layer. This rich decomposition allows novel image manipulations as demonstrated by our results."
]
} |
1902.03830 | 2926762703 | Intrinsic Image Decomposition (IID) is a challenging and interesting computer vision problem with various applications in several fields. We present novel semantic priors and an integrated approach for single image IID that involves analyzing image at three hierarchical context levels. Local context priors capture scene properties at each pixel within a small neighbourhood. Mid-level context priors encode object level semantics. Global context priors establish correspondences at the scene level. Our semantic priors are designed on both fixed and flexible regions, using selective search method and Convolutional Neural Network features. Our IID method is an iterative multistage optimization scheme and consists of two complementary formulations: @math smoothing for shading and @math sparsity for reflectance. Experiments and analysis of our method indicate the utility of our semantic priors and structured hierarchical analysis in an IID framework. We compare our method with other contemporary IID solutions and show results with lesser artifacts. Finally, we highlight that proper choice and encoding of prior knowledge can produce competitive results even when compared to end-to-end deep learning IID methods, signifying the importance of such priors. We believe that the insights and techniques presented in this paper would be useful in the future IID research. | A major challenge associated with IID research is lack of diverse large datasets and proper evaluation metrics . This arises mainly due to subjective nature of the problem and difficulty in collecting dense annotations. MIT intrinsic images dataset introduced by @cite_32 is limited to a handful of single object images on a black background. As-Realistic-As-Possible dataset by tries to capture complexity of natural scenes but is also not large enough for supervised training. Synthetic datasets like MPI Sintel by @cite_24 , provide dense annotation but lack sufficient diversity and complexity compared to the natural scenes. @cite_29 provide a large manually annotated dataset called Intrinsic Images in the Wild (IIW) but have only sparse relative annotations. This limits the utility of such datasets in learning based approaches which aim to work on complete scenes under unrestricted illumination and material property settings. | {
"cite_N": [
"@cite_24",
"@cite_29",
"@cite_32"
],
"mid": [
"1513100184",
"2076491823",
"2113404166"
],
"abstract": [
"Ground truth optical flow is difficult to measure in real scenes with natural motion. As a result, optical flow data sets are restricted in terms of size, complexity, and diversity, making optical flow algorithms difficult to train and test on realistic data. We introduce a new optical flow data set derived from the open source 3D animated short film Sintel. This data set has important features not present in the popular Middlebury flow evaluation: long sequences, large motions, specular reflections, motion blur, defocus blur, and atmospheric effects. Because the graphics data that generated the movie is open source, we are able to render scenes under conditions of varying complexity to evaluate where existing flow algorithms fail. We evaluate several recent optical flow algorithms and find that current highly-ranked methods on the Middlebury evaluation have difficulty with this more complex data set suggesting further research on optical flow estimation is needed. To validate the use of synthetic data, we compare the image- and flow-statistics of Sintel to those of real films and videos and show that they are similar. The data set, metrics, and evaluation website are publicly available.",
"Intrinsic image decomposition separates an image into a reflectance layer and a shading layer. Automatic intrinsic image decomposition remains a significant challenge, particularly for real-world scenes. Advances on this longstanding problem have been spurred by public datasets of ground truth data, such as the MIT Intrinsic Images dataset. However, the difficulty of acquiring ground truth data has meant that such datasets cover a small range of materials and objects. In contrast, real-world scenes contain a rich range of shapes and materials, lit by complex illumination. In this paper we introduce Intrinsic Images in the Wild, a large-scale, public dataset for evaluating intrinsic image decompositions of indoor scenes. We create this benchmark through millions of crowdsourced annotations of relative comparisons of material properties at pairs of points in each scene. Crowdsourcing enables a scalable approach to acquiring a large database, and uses the ability of humans to judge material comparisons, despite variations in illumination. Given our database, we develop a dense CRF-based intrinsic image algorithm for images in the wild that outperforms a range of state-of-the-art intrinsic image algorithms. Intrinsic image decomposition remains a challenging problem; we release our code and database publicly to support future research on this problem, available online at http: intrinsic.cs.cornell.edu .",
"The intrinsic image decomposition aims to retrieve “intrinsic” properties of an image, such as shading and reflectance. To make it possible to quantitatively compare different approaches to this problem in realistic settings, we present a ground-truth dataset of intrinsic image decompositions for a variety of real-world objects. For each object, we separate an image of it into three components: Lambertian shading, reflectance, and specularities. We use our dataset to quantitatively compare several existing algorithms; we hope that this dataset will serve as a means for evaluating future work on intrinsic images."
]
} |
1902.03830 | 2926762703 | Intrinsic Image Decomposition (IID) is a challenging and interesting computer vision problem with various applications in several fields. We present novel semantic priors and an integrated approach for single image IID that involves analyzing image at three hierarchical context levels. Local context priors capture scene properties at each pixel within a small neighbourhood. Mid-level context priors encode object level semantics. Global context priors establish correspondences at the scene level. Our semantic priors are designed on both fixed and flexible regions, using selective search method and Convolutional Neural Network features. Our IID method is an iterative multistage optimization scheme and consists of two complementary formulations: @math smoothing for shading and @math sparsity for reflectance. Experiments and analysis of our method indicate the utility of our semantic priors and structured hierarchical analysis in an IID framework. We compare our method with other contemporary IID solutions and show results with lesser artifacts. Finally, we highlight that proper choice and encoding of prior knowledge can produce competitive results even when compared to end-to-end deep learning IID methods, signifying the importance of such priors. We believe that the insights and techniques presented in this paper would be useful in the future IID research. | Yet another challenge in IID research is lack of proper evaluation metric which reflects both quantitative and qualitative performance. Local Mean Square Error (LMSE) and Structural Similarity Index Metric (SSIM) are used for synthetic scenes but these metrics require dense ground truth annotations. @cite_29 suggest a new performance evaluation metric based on their IIW dataset: Weighted Human Disagreement Rate (WHDR). WHDR gives relative error rate within a given threshold based on the sparse annotations in the IIW dataset. Lack of a proper evaluation metric which could properly evaluate results both perceptually and objectively for all application scenarios, makes comparison between different IID solutions a difficult task. | {
"cite_N": [
"@cite_29"
],
"mid": [
"2076491823"
],
"abstract": [
"Intrinsic image decomposition separates an image into a reflectance layer and a shading layer. Automatic intrinsic image decomposition remains a significant challenge, particularly for real-world scenes. Advances on this longstanding problem have been spurred by public datasets of ground truth data, such as the MIT Intrinsic Images dataset. However, the difficulty of acquiring ground truth data has meant that such datasets cover a small range of materials and objects. In contrast, real-world scenes contain a rich range of shapes and materials, lit by complex illumination. In this paper we introduce Intrinsic Images in the Wild, a large-scale, public dataset for evaluating intrinsic image decompositions of indoor scenes. We create this benchmark through millions of crowdsourced annotations of relative comparisons of material properties at pairs of points in each scene. Crowdsourcing enables a scalable approach to acquiring a large database, and uses the ability of humans to judge material comparisons, despite variations in illumination. Given our database, we develop a dense CRF-based intrinsic image algorithm for images in the wild that outperforms a range of state-of-the-art intrinsic image algorithms. Intrinsic image decomposition remains a challenging problem; we release our code and database publicly to support future research on this problem, available online at http: intrinsic.cs.cornell.edu ."
]
} |
1902.03830 | 2926762703 | Intrinsic Image Decomposition (IID) is a challenging and interesting computer vision problem with various applications in several fields. We present novel semantic priors and an integrated approach for single image IID that involves analyzing image at three hierarchical context levels. Local context priors capture scene properties at each pixel within a small neighbourhood. Mid-level context priors encode object level semantics. Global context priors establish correspondences at the scene level. Our semantic priors are designed on both fixed and flexible regions, using selective search method and Convolutional Neural Network features. Our IID method is an iterative multistage optimization scheme and consists of two complementary formulations: @math smoothing for shading and @math sparsity for reflectance. Experiments and analysis of our method indicate the utility of our semantic priors and structured hierarchical analysis in an IID framework. We compare our method with other contemporary IID solutions and show results with lesser artifacts. Finally, we highlight that proper choice and encoding of prior knowledge can produce competitive results even when compared to end-to-end deep learning IID methods, signifying the importance of such priors. We believe that the insights and techniques presented in this paper would be useful in the future IID research. | These issues concerning datasets and performance evaluation, along with it ill-defined nature, make IID a challenging problem to solve using supervised learning frameworks. On the other hand, several older IID methods were unsupervised in nature. @cite_22 , @cite_1 and @cite_25 relied on intelligently designed priors and designed their method as optimization schemes. Such earlier models take advantage of prior knowledge, but either work in restricted settings based upon the assumptions in the models or have scope for performance improvement compared to deep learning based schemes. CNNs have been widely used in computer vision and machine learning literature as black box feature extractor . @cite_31 directly use pre-trained CNNs as a feature extractor and prove the generality and cross domain applicability of such features on varied tasks like scene recognition, fine-grained recognition and domain adaptation. Along similar lines, @cite_11 and @cite_9 also use these features on increasingly different tasks and datasets, highlighting their task agnostic characteristics. | {
"cite_N": [
"@cite_22",
"@cite_9",
"@cite_1",
"@cite_31",
"@cite_25",
"@cite_11"
],
"mid": [
"2136748901",
"2949667497",
"2119523460",
"2953360861",
"2057552653",
"2953391683"
],
"abstract": [
"Intrinsic images are a useful midlevel description of scenes proposed by H.G. Barrow and J.M. Tenenbaum (1978). An image is de-composed into two images: a reflectance image and an illumination image. Finding such a decomposition remains a difficult problem in computer vision. We focus on a slightly, easier problem: given a sequence of T images where the reflectance is constant and the illumination changes, can we recover T illumination images and a single reflectance image? We show that this problem is still imposed and suggest approaching it as a maximum-likelihood estimation problem. Following recent work on the statistics of natural images, we use a prior that assumes that illumination images will give rise to sparse filter outputs. We show that this leads to a simple, novel algorithm for recovering reflectance images. We illustrate the algorithm's performance on real and synthetic image sequences.",
"Many deep neural networks trained on natural images exhibit a curious phenomenon in common: on the first layer they learn features similar to Gabor filters and color blobs. Such first-layer features appear not to be specific to a particular dataset or task, but general in that they are applicable to many datasets and tasks. Features must eventually transition from general to specific by the last layer of the network, but this transition has not been studied extensively. In this paper we experimentally quantify the generality versus specificity of neurons in each layer of a deep convolutional neural network and report a few surprising results. Transferability is negatively affected by two distinct issues: (1) the specialization of higher layer neurons to their original task at the expense of performance on the target task, which was expected, and (2) optimization difficulties related to splitting networks between co-adapted neurons, which was not expected. In an example network trained on ImageNet, we demonstrate that either of these two issues may dominate, depending on whether features are transferred from the bottom, middle, or top of the network. We also document that the transferability of features decreases as the distance between the base task and target task increases, but that transferring features even from distant tasks can be better than using random features. A final surprising result is that initializing a network with transferred features from almost any number of layers can produce a boost to generalization that lingers even after fine-tuning to the target dataset.",
"We address the challenging task of decoupling material properties from lighting properties given a single image. In the last two decades virtually all works have concentrated on exploiting edge information to address this problem. We take a different route by introducing a new prior on reflectance, that models reflectance values as being drawn from a sparse set of basis colors. This results in a Random Field model with global, latent variables (basis colors) and pixel-accurate output reflectance values. We show that without edge information high-quality results can be achieved, that are on par with methods exploiting this source of information. Finally, we are able to improve on state-of-the-art results by integrating edge information into our model. We believe that our new approach is an excellent starting point for future developments in this field.",
"We evaluate whether features extracted from the activation of a deep convolutional network trained in a fully supervised fashion on a large, fixed set of object recognition tasks can be re-purposed to novel generic tasks. Our generic tasks may differ significantly from the originally trained tasks and there may be insufficient labeled or unlabeled data to conventionally train or adapt a deep architecture to the new tasks. We investigate and visualize the semantic clustering of deep convolutional features with respect to a variety of such tasks, including scene recognition, domain adaptation, and fine-grained recognition challenges. We compare the efficacy of relying on various network levels to define a fixed feature, and report novel results that significantly outperform the state-of-the-art on several important vision challenges. We are releasing DeCAF, an open-source implementation of these deep convolutional activation features, along with all associated network parameters to enable vision researchers to be able to conduct experimentation with deep representations across a range of visual concept learning paradigms.",
"We introduce a method to compute intrinsic images for a multiview set of outdoor photos with cast shadows, taken under the same lighting. We use an automatic 3D reconstruction from these photos and the sun direction as input and decompose each image into reflectance and shading layers, despite the inaccuracies and missing data of the 3D model. Our approach is based on two key ideas. First, we progressively improve the accuracy of the parameters of our image formation model by performing iterative estimation and combining 3D lighting simulation with 2D image optimization methods. Second, we use the image formation model to express reflectance as a function of discrete visibility values for shadow and light, which allows to introduce a robust visibility classifier for pairs of points in a scene. This classifier is used for shadow labeling, allowing to compute high-quality reflectance and shading layers. Our multiview intrinsic decomposition is of sufficient quality to allow relighting of the input images. We create shadow-caster geometry which preserves shadow silhouettes and, using the intrinsic layers, we can perform multiview relighting with moving cast shadows. We present results on several multiview datasets, and show how it is now possible to perform image-based rendering with changing illumination conditions.",
"Recent results indicate that the generic descriptors extracted from the convolutional neural networks are very powerful. This paper adds to the mounting evidence that this is indeed the case. We report on a series of experiments conducted for different recognition tasks using the publicly available code and model of the network which was trained to perform object classification on ILSVRC13. We use features extracted from the network as a generic image representation to tackle the diverse range of recognition tasks of object image classification, scene recognition, fine grained recognition, attribute detection and image retrieval applied to a diverse set of datasets. We selected these tasks and datasets as they gradually move further away from the original task and data the network was trained to solve. Astonishingly, we report consistent superior results compared to the highly tuned state-of-the-art systems in all the visual classification tasks on various datasets. For instance retrieval it consistently outperforms low memory footprint methods except for sculptures dataset. The results are achieved using a linear SVM classifier (or @math distance in case of retrieval) applied to a feature representation of size 4096 extracted from a layer in the net. The representations are further modified using simple augmentation techniques e.g. jittering. The results strongly suggest that features obtained from deep learning with convolutional nets should be the primary candidate in most visual recognition tasks."
]
} |
1902.03830 | 2926762703 | Intrinsic Image Decomposition (IID) is a challenging and interesting computer vision problem with various applications in several fields. We present novel semantic priors and an integrated approach for single image IID that involves analyzing image at three hierarchical context levels. Local context priors capture scene properties at each pixel within a small neighbourhood. Mid-level context priors encode object level semantics. Global context priors establish correspondences at the scene level. Our semantic priors are designed on both fixed and flexible regions, using selective search method and Convolutional Neural Network features. Our IID method is an iterative multistage optimization scheme and consists of two complementary formulations: @math smoothing for shading and @math sparsity for reflectance. Experiments and analysis of our method indicate the utility of our semantic priors and structured hierarchical analysis in an IID framework. We compare our method with other contemporary IID solutions and show results with lesser artifacts. Finally, we highlight that proper choice and encoding of prior knowledge can produce competitive results even when compared to end-to-end deep learning IID methods, signifying the importance of such priors. We believe that the insights and techniques presented in this paper would be useful in the future IID research. | In our model, we absorb the advantages of both the supervised and unsupervised approaches by combining the generality of supervised deep learning methods with prior domain knowledge. We employ an off-the-shelf' pre-trained deep neural network as a black-box to obtain generic features. Additionally we also use an unsupervised method to provide yet another set of semantic features. We use both these features to introduce new context priors in an unsupervised optimization algorithm by posing it as a standard total variation optimization problem @cite_23 . Recently great improvements have been made in IID with the use of supervised methods primarily utilizing new large new training datasets such as: time lapse dataset , multi-illumination dataset and synthetic datasets . Bigger datasets improve the performance of supervised learning systems, but we can improve these solutions by explicitly encoding problem specific insights. We believe our IID specific semantic priors, encode crucial domain insights and would further help such methods. | {
"cite_N": [
"@cite_23"
],
"mid": [
"2142058898"
],
"abstract": [
"The class of L1-regularized optimization problems has received much attention recently because of the introduction of “compressed sensing,” which allows images and signals to be reconstructed from small amounts of data. Despite this recent attention, many L1-regularized problems still remain difficult to solve, or require techniques that are very problem-specific. In this paper, we show that Bregman iteration can be used to solve a wide variety of constrained optimization problems. Using this technique, we propose a “split Bregman” method, which can solve a very broad class of L1-regularized problems. We apply this technique to the Rudin-Osher-Fatemi functional for image denoising and to a compressed sensing problem that arises in magnetic resonance imaging."
]
} |
1902.03842 | 2935143371 | This paper uses robust statistics and curvelet transform to learn a general-purpose no-reference (NR) image quality assessment (IQA) model. The new approach, here called M1, competes with the Curvelet Quality Assessment proposed in 2014 (Curvelet2014). The central idea is to use descriptors based on robust statistics to extract features and predict the human opinion about degraded images. To show the consistency of the method the model is tested with 3 different datasets, LIVE IQA, TID2013 and CSIQ. To test evaluation, it is used the Wilcoxon test to verify the statistical significance of results and promote an accurate comparison between new model M1 and Curvelet2014. The results show a gain when robust statistics are used as descriptor. | The first use of FDCT in NR IQA proposals was made by @cite_2 , using only the finest layer for feature extraction. The architecture 2-stages SVM was introduced by Moorthy and Bovik @cite_11 applied to features extracted from natural scenes. This architecture was used in by @cite_10 @cite_8 @cite_17 @cite_4 with different features. | {
"cite_N": [
"@cite_4",
"@cite_8",
"@cite_17",
"@cite_2",
"@cite_10",
"@cite_11"
],
"mid": [
"2162692770",
"2774447305",
"1982471090",
"1989735895",
"1537865522",
"2163370434"
],
"abstract": [
"We develop an efficient general-purpose blind no-reference image quality assessment (IQA) algorithm using a natural scene statistics (NSS) model of discrete cosine transform (DCT) coefficients. The algorithm is computationally appealing, given the availability of platforms optimized for DCT computation. The approach relies on a simple Bayesian inference model to predict image quality scores given certain extracted features. The features are based on an NSS model of the image DCT coefficients. The estimated parameters of the model are utilized to form features that are indicative of perceptual quality. These features are used in a simple Bayesian inference approach to predict quality scores. The resulting algorithm, which we name BLIINDS-II, requires minimal training and adopts a simple probabilistic model for score prediction. Given the extracted features from a test image, the quality score that maximizes the probability of the empirically determined inference model is chosen as the predicted quality score of that image. When tested on the LIVE IQA database, BLIINDS-II is shown to correlate highly with human judgments of quality, at a level that is competitive with the popular SSIM index.",
"Contrast is a very important characteristic for visual perception of image quality. Some No-Reference Image Quality Assessment Algorithm NR-IQA metrics for Contrast-Distorted Images (CDI) have been proposed in the literature, e.g. Reduced-reference Image Quality Metric for Contrast-changed images (RIQMC) and NR-IQA for Contrast-Distorted Images (NR-IQACDI). Here, we intend to improve the assessment results of images available in databases such as TID2013 and CSIQ. Most of the NR-IQA metrics (e.g. NR-IQACDI) designed for CDI adopt features available in the spatial domain. This paper proposes to compliment it with feature in Curvelet domain which is powerful in capturing multiscale and multidirectional information in an image. We employed the Natural Scene Statistics (NSS) features in Curvelet domain originally recommended by (2014) which were found useful in the assessment of the quality of image distorted by compression, noise and blurring. Experiments were then conducted to assess the effect of incorporating these NSS features. The experimental results based on K-fold cross validation (K ranged from 2 to 10) and statistical test showed that the performance of NRIQACDI was improved. Future works include improvements of NRIQACDI, exploration of feature fusion methods and using a suitable feature selection method.",
"We propose a natural scene statistic-based distortion-generic blind no-reference (NR) image quality assessment (IQA) model that operates in the spatial domain. The new model, dubbed blind referenceless image spatial quality evaluator (BRISQUE) does not compute distortion-specific features, such as ringing, blur, or blocking, but instead uses scene statistics of locally normalized luminance coefficients to quantify possible losses of “naturalness” in the image due to the presence of distortions, thereby leading to a holistic measure of quality. The underlying features used derive from the empirical distribution of locally normalized luminances and products of locally normalized luminances under a spatial natural scene statistic model. No transformation to another coordinate frame (DCT, wavelet, etc.) is required, distinguishing it from prior NR IQA approaches. Despite its simplicity, we are able to show that BRISQUE is statistically better than the full-reference peak signal-to-noise ratio and the structural similarity index, and is highly competitive with respect to all present-day distortion-generic NR IQA algorithms. BRISQUE has very low computational complexity, making it well suited for real time applications. BRISQUE features may be used for distortion-identification as well. To illustrate a new practical application of BRISQUE, we describe how a nonblind image denoising algorithm can be augmented with BRISQUE in order to perform blind image denoising. Results show that BRISQUE augmentation leads to performance improvements over state-of-the-art methods. A software release of BRISQUE is available online: http: live.ece.utexas.edu research quality BRISQUE_release.zip for public use and evaluation.",
"In this paper, we propose a new general Quality Assessment method based on the curvelet transform, called Curvelet No-Reference (CNR) model, which can estimate levels of noise, blur and JPEG 2000 compression of natural images. The peak coordinate of the curvelet coefficient histogram occupies distinctive regions depending on how the image was modified from the original. During training, we associate peak positions with known filter levels. In the prediction stage, the filter levels of new images are estimated from the training data, with no access to the reference images. We tested CNR both on our own image dataset and on LIVE [4]. Results demonstrate that CNR does a better job at predicting noise and blur levels among several methods, including SSIM [1] and PSNR. We also present an accelerated version of CNR that does not sacrifice prediction accuracy on natural images.",
"We develop an efficient general-purpose blind no-reference image quality assessment (IQA) algorithm that utilizes curvelet domain features of phase congruency values and local spectral entropy values in distorted images. A 2-stage framework of distortion classification followed by quality assessment is used for mapping feature vectors to prediction scores. We utilize a support vector machine (SVM) to train an image distortion and quality prediction model. The resulting algorithm which we name Phase Congruency and Spectral Entropy based Quality (PCSEQ) index is capable of assessing the quality of distorted images across multiple distortion categories. We explain the advantages of phase congruency features and spectral entropy features. We also thoroughly test the algorithm on the LIVE IQA databse and find that PCSEQ correlates well with human judgments of quality. It is superior to the full-reference (FR) IQA algorithm SSIM and several top-performance no-reference (NR) IQA methods such as DIIVINE and SSEQ. We also tested PCSEQ on the TID2008 database to ascertain whether it has performance that is database independent.",
"Present day no-reference no-reference image quality assessment (NR IQA) algorithms usually assume that the distortion affecting the image is known. This is a limiting assumption for practical applications, since in a majority of cases the distortions in the image are unknown. We propose a new two-step framework for no-reference image quality assessment based on natural scene statistics (NSS). Once trained, the framework does not require any knowledge of the distorting process and the framework is modular in that it can be extended to any number of distortions. We describe the framework for blind image quality assessment and a version of this framework-the blind image quality index (BIQI) is evaluated on the LIVE image quality assessment database. A software release of BIQI has been made available online: http: live.ece.utexas.edu research quality BIQI_release.zip."
]
} |
1902.03842 | 2935143371 | This paper uses robust statistics and curvelet transform to learn a general-purpose no-reference (NR) image quality assessment (IQA) model. The new approach, here called M1, competes with the Curvelet Quality Assessment proposed in 2014 (Curvelet2014). The central idea is to use descriptors based on robust statistics to extract features and predict the human opinion about degraded images. To show the consistency of the method the model is tested with 3 different datasets, LIVE IQA, TID2013 and CSIQ. To test evaluation, it is used the Wilcoxon test to verify the statistical significance of results and promote an accurate comparison between new model M1 and Curvelet2014. The results show a gain when robust statistics are used as descriptor. | The use of FDCT with 2-stages SVM was proposed by @cite_5 . @cite_10 proposed an NR IQA method that uses the entropy of the Curvelet transform coefficients, also associated with the 2-stages SVM. Ahmed e Der @cite_8 used the method proposed in @cite_5 with the spatial features for study of contrast in images. Recently @cite_12 use the model Curvelet2014 @cite_5 to quantify the quality of images associated with historical documents. | {
"cite_N": [
"@cite_5",
"@cite_10",
"@cite_12",
"@cite_8"
],
"mid": [
"2008051965",
"1537865522",
"2767619925",
"2774447305"
],
"abstract": [
"Abstract We study the efficacy of utilizing a powerful image descriptor, the curvelet transform, to learn a no-reference (NR) image quality assessment (IQA) model. A set of statistical features are extracted from a computed image curvelet representation, including the coordinates of the maxima of the log-histograms of the curvelet coefficients values, and the energy distributions of both orientation and scale in the curvelet domain. Our results indicate that these features are sensitive to the presence and severity of image distortion. Operating within a 2-stage framework of distortion classification followed by quality assessment, we train an image distortion and quality prediction engine using a support vector machine (SVM). The resulting algorithm, dubbed CurveletQA for short, was tested on the LIVE IQA database and compared to state-of-the-art NR FR IQA algorithms. We found that CurveletQA correlates well with human subjective opinions of image quality, delivering performance that is competitive with popular full-reference (FR) IQA algorithms such as SSIM, and with top-performing NR IQA models. At the same time, CurveletQA has a relatively low complexity.",
"We develop an efficient general-purpose blind no-reference image quality assessment (IQA) algorithm that utilizes curvelet domain features of phase congruency values and local spectral entropy values in distorted images. A 2-stage framework of distortion classification followed by quality assessment is used for mapping feature vectors to prediction scores. We utilize a support vector machine (SVM) to train an image distortion and quality prediction model. The resulting algorithm which we name Phase Congruency and Spectral Entropy based Quality (PCSEQ) index is capable of assessing the quality of distorted images across multiple distortion categories. We explain the advantages of phase congruency features and spectral entropy features. We also thoroughly test the algorithm on the LIVE IQA databse and find that PCSEQ correlates well with human judgments of quality. It is superior to the full-reference (FR) IQA algorithm SSIM and several top-performance no-reference (NR) IQA methods such as DIIVINE and SSEQ. We also tested PCSEQ on the TID2008 database to ascertain whether it has performance that is database independent.",
"Abstract The huge amount of degraded documents stored in libraries and archives around the world needs automatic procedures of enhancement, classification, transliteration, etc. While high-quality images of these documents are in general easy to be captured, the amount of damage these documents contain before imaging is unknown. It is highly desirable to measure the severity of degradation that each document image contains. The degradation assessment can be used in tuning parameters of processing algorithms, selecting the proper algorithm, finding damaged or exceptional documents, among other applications. In this paper, the first dataset of degraded document images along with the human opinion scores for each document image is introduced in order to evaluate the image quality assessment metrics on historical document images. In this research, human judgments on the overall quality of the document image are used instead of the previously used OCR performance. Also, we propose an objective no reference quality metric based on the statistics of the mean subtracted contrast normalized (MSCN) coefficients computed from segmented layers of each document image. The segmentation into four layers of foreground and background is done on the basis of an analysis of the log-Gabor filters. This segmentation is based on the assumption that the sensitivity of the human visual system (HVS) is different at the locations of text and non-text. Experimental results show that the proposed metric has comparable or better performance than the state-of-the-art metrics, while it has a moderate complexity. The developed dataset as well as the Matlab source code of the proposed metric is available at http: www.synchromedia.ca system files VDIQA.zip .",
"Contrast is a very important characteristic for visual perception of image quality. Some No-Reference Image Quality Assessment Algorithm NR-IQA metrics for Contrast-Distorted Images (CDI) have been proposed in the literature, e.g. Reduced-reference Image Quality Metric for Contrast-changed images (RIQMC) and NR-IQA for Contrast-Distorted Images (NR-IQACDI). Here, we intend to improve the assessment results of images available in databases such as TID2013 and CSIQ. Most of the NR-IQA metrics (e.g. NR-IQACDI) designed for CDI adopt features available in the spatial domain. This paper proposes to compliment it with feature in Curvelet domain which is powerful in capturing multiscale and multidirectional information in an image. We employed the Natural Scene Statistics (NSS) features in Curvelet domain originally recommended by (2014) which were found useful in the assessment of the quality of image distorted by compression, noise and blurring. Experiments were then conducted to assess the effect of incorporating these NSS features. The experimental results based on K-fold cross validation (K ranged from 2 to 10) and statistical test showed that the performance of NRIQACDI was improved. Future works include improvements of NRIQACDI, exploration of feature fusion methods and using a suitable feature selection method."
]
} |
1902.03842 | 2935143371 | This paper uses robust statistics and curvelet transform to learn a general-purpose no-reference (NR) image quality assessment (IQA) model. The new approach, here called M1, competes with the Curvelet Quality Assessment proposed in 2014 (Curvelet2014). The central idea is to use descriptors based on robust statistics to extract features and predict the human opinion about degraded images. To show the consistency of the method the model is tested with 3 different datasets, LIVE IQA, TID2013 and CSIQ. To test evaluation, it is used the Wilcoxon test to verify the statistical significance of results and promote an accurate comparison between new model M1 and Curvelet2014. The results show a gain when robust statistics are used as descriptor. | Examples of NR IQA work using non-parametric statistics associated with the Curvelet transform can be found in @cite_2 @cite_10 . On the other hand, in the literature review, the authors did not find any work in NR IQA that made use of robust statistical tools to extract features from the Curvelets space. | {
"cite_N": [
"@cite_10",
"@cite_2"
],
"mid": [
"1537865522",
"1989735895"
],
"abstract": [
"We develop an efficient general-purpose blind no-reference image quality assessment (IQA) algorithm that utilizes curvelet domain features of phase congruency values and local spectral entropy values in distorted images. A 2-stage framework of distortion classification followed by quality assessment is used for mapping feature vectors to prediction scores. We utilize a support vector machine (SVM) to train an image distortion and quality prediction model. The resulting algorithm which we name Phase Congruency and Spectral Entropy based Quality (PCSEQ) index is capable of assessing the quality of distorted images across multiple distortion categories. We explain the advantages of phase congruency features and spectral entropy features. We also thoroughly test the algorithm on the LIVE IQA databse and find that PCSEQ correlates well with human judgments of quality. It is superior to the full-reference (FR) IQA algorithm SSIM and several top-performance no-reference (NR) IQA methods such as DIIVINE and SSEQ. We also tested PCSEQ on the TID2008 database to ascertain whether it has performance that is database independent.",
"In this paper, we propose a new general Quality Assessment method based on the curvelet transform, called Curvelet No-Reference (CNR) model, which can estimate levels of noise, blur and JPEG 2000 compression of natural images. The peak coordinate of the curvelet coefficient histogram occupies distinctive regions depending on how the image was modified from the original. During training, we associate peak positions with known filter levels. In the prediction stage, the filter levels of new images are estimated from the training data, with no access to the reference images. We tested CNR both on our own image dataset and on LIVE [4]. Results demonstrate that CNR does a better job at predicting noise and blur levels among several methods, including SSIM [1] and PSNR. We also present an accelerated version of CNR that does not sacrifice prediction accuracy on natural images."
]
} |
1902.03274 | 2919104072 | Suffix trees are a fundamental data structure in stringology, but their space usage, though linear, is an important problem for its applications. We design and implement a new compressed suffix tree targeted to highly repetitive texts, such as large genomic collections of the same species. Our suffix tree tree builds on Block Trees, a recent Lempel-Ziv-bounded data structure that captures the repetitiveness of its input. We use Block Trees to compress the topology of the suffix tree, and augment the Block Tree nodes with data that speeds up suffix tree navigation. Our compressed suffix tree is slightly larger than previous repetition-aware suffix trees based on grammars, but outperforms them in time, often by orders of magnitude. The component that represents the tree topology achieves a speed comparable to that of general-purpose compressed trees, while using 2.3--10 times less space, and might be of interest in other scenarios. | The suffix array @cite_23 @math of a text @math is a permutation of @math such that @math is the starting position of the @math th suffix in increasing lexicographical order. Note that the leaves descending from a suffix tree node span a range of suffixes in @math . | {
"cite_N": [
"@cite_23"
],
"mid": [
"2158874082"
],
"abstract": [
"A new and conceptually simple data structure, called a suffix array, for on-line string searches is introduced in this paper. Constructing and querying suffix arrays is reduced to a sort and search paradigm that employs novel algorithms. The main advantage of suffix arrays over suffix trees is that, in practice, they use three to five times less space. From a complexity standpoint, suffix arrays permit on-line string searches of the type, “Is W a substring of A?” to be answered in time @math , where P is the length of W and N is the length of A, which is competitive with (and in some cases slightly better than) suffix trees. The only drawback is that in those instances where the underlying alphabet is finite and small, suffix trees can be constructed in @math time in the worst case, versus @math time for suffix arrays. However, an augmented algorithm is given that, regardless of the alphabet size, constructs suffix arrays in @math expected time, albeit with lesser space efficiency. It is ..."
]
} |
1902.03274 | 2919104072 | Suffix trees are a fundamental data structure in stringology, but their space usage, though linear, is an important problem for its applications. We design and implement a new compressed suffix tree targeted to highly repetitive texts, such as large genomic collections of the same species. Our suffix tree tree builds on Block Trees, a recent Lempel-Ziv-bounded data structure that captures the repetitiveness of its input. We use Block Trees to compress the topology of the suffix tree, and augment the Block Tree nodes with data that speeds up suffix tree navigation. Our compressed suffix tree is slightly larger than previous repetition-aware suffix trees based on grammars, but outperforms them in time, often by orders of magnitude. The component that represents the tree topology achieves a speed comparable to that of general-purpose compressed trees, while using 2.3--10 times less space, and might be of interest in other scenarios. | The function @math is the length of the longest common prefix (lcp) of strings @math and @math . The LCP array @cite_23 , @math , is defined as @math and @math for all @math , that is, it stores the lengths of the longest common prefixes between lexicographically consecutive suffixes of @math . | {
"cite_N": [
"@cite_23"
],
"mid": [
"2158874082"
],
"abstract": [
"A new and conceptually simple data structure, called a suffix array, for on-line string searches is introduced in this paper. Constructing and querying suffix arrays is reduced to a sort and search paradigm that employs novel algorithms. The main advantage of suffix arrays over suffix trees is that, in practice, they use three to five times less space. From a complexity standpoint, suffix arrays permit on-line string searches of the type, “Is W a substring of A?” to be answered in time @math , where P is the length of W and N is the length of A, which is competitive with (and in some cases slightly better than) suffix trees. The only drawback is that in those instances where the underlying alphabet is finite and small, suffix trees can be constructed in @math time in the worst case, versus @math time for suffix arrays. However, an augmented algorithm is given that, regardless of the alphabet size, constructs suffix arrays in @math expected time, albeit with lesser space efficiency. It is ..."
]
} |
1902.03274 | 2919104072 | Suffix trees are a fundamental data structure in stringology, but their space usage, though linear, is an important problem for its applications. We design and implement a new compressed suffix tree targeted to highly repetitive texts, such as large genomic collections of the same species. Our suffix tree tree builds on Block Trees, a recent Lempel-Ziv-bounded data structure that captures the repetitiveness of its input. We use Block Trees to compress the topology of the suffix tree, and augment the Block Tree nodes with data that speeds up suffix tree navigation. Our compressed suffix tree is slightly larger than previous repetition-aware suffix trees based on grammars, but outperforms them in time, often by orders of magnitude. The component that represents the tree topology achieves a speed comparable to that of general-purpose compressed trees, while using 2.3--10 times less space, and might be of interest in other scenarios. | A balanced parentheses (BP) representation (there are others @cite_21 ) of the topology of an ordinal tree @math of @math nodes is a binary sequence (or bitvector) @math built as follows: we traverse @math in preorder, writing an opening parenthesis (a bit 1) when we first arrive at a node, and a closing one (a bit 0) when we leave its subtree. For example, a leaf looks like 10''. The following primitives can be defined on @math : | {
"cite_N": [
"@cite_21"
],
"mid": [
"165200395"
],
"abstract": [
"We survey succinct representations of ordinal, or rooted, ordered trees. Succinct representations use space that is close to the appropriate information-theoretic minimum, but support operations on the tree rapidly, usually in O(1) time."
]
} |
1902.03274 | 2919104072 | Suffix trees are a fundamental data structure in stringology, but their space usage, though linear, is an important problem for its applications. We design and implement a new compressed suffix tree targeted to highly repetitive texts, such as large genomic collections of the same species. Our suffix tree tree builds on Block Trees, a recent Lempel-Ziv-bounded data structure that captures the repetitiveness of its input. We use Block Trees to compress the topology of the suffix tree, and augment the Block Tree nodes with data that speeds up suffix tree navigation. Our compressed suffix tree is slightly larger than previous repetition-aware suffix trees based on grammars, but outperforms them in time, often by orders of magnitude. The component that represents the tree topology achieves a speed comparable to that of general-purpose compressed trees, while using 2.3--10 times less space, and might be of interest in other scenarios. | A milestone in the area was the emergence of Compressed Suffix Arrays (CSAs) @cite_26 , which using space proportional to that of the compressed sequence managed to answer access queries to the original suffix array and its inverse (i.e., return any @math and @math ), to the indexed sequence (i.e., return any @math ), and access to a novel array, @math , which lets us move from a text suffix @math to the next one, @math , yet indexing the suffixes by their lexicographic rank, @math . This function plays an important role in the design of CSTs, as we see next. | {
"cite_N": [
"@cite_26"
],
"mid": [
"2088386938"
],
"abstract": [
"Full-text indexes provide fast substring search over large text collections. A serious problem of these indexes has traditionally been their space consumption. A recent trend is to develop indexes that exploit the compressibility of the text, so that their size is a function of the compressed text length. This concept has evolved into self-indexes, which in addition contain enough information to reproduce any text portion, so they replace the text. The exciting possibility of an index that takes space close to that of the compressed text, replaces it, and in addition provides fast search over it, has triggered a wealth of activity and produced surprising results in a very short time, which radically changed the status of this area in less than 5 years. The most successful indexes nowadays are able to obtain almost optimal space and search time simultaneously. In this article we present the main concepts underlying (compressed) self-indexes. We explain the relationship between text entropy and regularities that show up in index structures and permit compressing them. Then we cover the most relevant self-indexes, focusing on how they exploit text compressibility to achieve compact structures that can efficiently solve various search problems. Our aim is to give the background to understand and follow the developments in this area."
]
} |
1902.03155 | 2919816369 | In this paper, we introduce BINet, a neural network architecture for real-time multi-perspective anomaly detection in business process event logs. BINet is designed to handle both the control flow and the data perspective of a business process. Additionally, we propose a set of heuristics for setting the threshold of an anomaly detection algorithm automatically. We demonstrate that BINet can be used to detect anomalies in event logs not only on a case level but also on event attribute level. Finally, we demonstrate that a simple set of rules can be used to utilize the output of BINet for anomaly classification. We compare BINet to eight other state-of-the-art anomaly detection algorithms and evaluate their performance on an elaborate data corpus of 29 synthetic and 15 real-life event logs. BINet outperforms all other methods both on the synthetic as well as on the real-life datasets. | A more recent publication proposes the use of likelihood graphs to analyze business process behavior @cite_16 . This method includes important characteristics of the process itself by including the event attributes as part of an extended likelihood graph. However, this method relies on a discrete order in which the attributes are connected to the graph, which may introduce a bias towards certain attributes. Furthermore, the same activities are mapped to the same node in the likelihood graph, thereby assigning a single probability distribution to each activity. In other words, control flow dependencies cannot be modeled by the likelihood graph, because the probability distribution of attributes following an activity does not depend on the history of events. | {
"cite_N": [
"@cite_16"
],
"mid": [
"2538859255"
],
"abstract": [
"Ensuring anomaly-free process model executions is crucial in order to prevent fraud and security breaches. Existing anomaly detection approaches focus on the control flow, point anomalies, and struggle with false positives in the case of unexpected events. By contrast, this paper proposes an anomaly detection approach that incorporates perspectives that go beyond the control flow, such as, time and resources (i.e., to detect contextual anomalies). In addition, it is capable of dealing with unexpected process model execution events: not every unexpected event is immediately detected as anomalous, but based on a certain likelihood of occurrence, hence reducing the number of false positives. Finally, multiple events are analyzed in a combined manner in order to detect collective anomalies. The performance and applicability of the overall approach are evaluated by means of a prototypical implementation along and based on real life process execution logs from multiple domains."
]
} |
1902.03155 | 2919816369 | In this paper, we introduce BINet, a neural network architecture for real-time multi-perspective anomaly detection in business process event logs. BINet is designed to handle both the control flow and the data perspective of a business process. Additionally, we propose a set of heuristics for setting the threshold of an anomaly detection algorithm automatically. We demonstrate that BINet can be used to detect anomalies in event logs not only on a case level but also on event attribute level. Finally, we demonstrate that a simple set of rules can be used to utilize the output of BINet for anomaly classification. We compare BINet to eight other state-of-the-art anomaly detection algorithms and evaluate their performance on an elaborate data corpus of 29 synthetic and 15 real-life event logs. BINet outperforms all other methods both on the synthetic as well as on the real-life datasets. | A review of traditional anomaly detection methodology can be found in @cite_3 . Here, the authors describe and compare many methods that have been proposed over the last decades. Another elaborate summary of anomaly detection in discrete sequences is given by Chandola et ,al. in @cite_17 . The authors differentiate between five different basic methods for novelty detection: probabilistic, distance-based, reconstruction-based, domain-based, and information-theoretic novelty detection. | {
"cite_N": [
"@cite_3",
"@cite_17"
],
"mid": [
"2115627867",
"2038819732"
],
"abstract": [
"Novelty detection is the task of classifying test data that differ in some respect from the data that are available during training. This may be seen as ''one-class classification'', in which a model is constructed to describe ''normal'' training data. The novelty detection approach is typically used when the quantity of available ''abnormal'' data is insufficient to construct explicit models for non-normal classes. Application includes inference in datasets from critical systems, where the quantity of available normal data is very large, such that ''normality'' may be accurately modelled. In this review we aim to provide an updated and structured investigation of novelty detection research papers that have appeared in the machine learning literature during the last decade.",
"This survey attempts to provide a comprehensive and structured overview of the existing research for the problem of detecting anomalies in discrete symbolic sequences. The objective is to provide a global understanding of the sequence anomaly detection problem and how existing techniques relate to each other. The key contribution of this survey is the classification of the existing research into three distinct categories, based on the problem formulation that they are trying to solve. These problem formulations are: 1) identifying anomalous sequences with respect to a database of normal sequences; 2) identifying an anomalous subsequence within a long sequence; and 3) identifying a pattern in a sequence whose frequency of occurrence is anomalous. We show how each of these problem formulations is characteristically distinct from each other and discuss their relevance in various application domains. We review techniques from many disparate and disconnected application domains that address each of these formulations. Within each problem formulation, we group techniques into categories based on the nature of the underlying algorithm. For each category, we provide a basic anomaly detection technique, and show how the existing techniques are variants of the basic technique. This approach shows how different techniques within a category are related or different from each other. Our categorization reveals new variants and combinations that have not been investigated before for anomaly detection. We also provide a discussion of relative strengths and weaknesses of different techniques. We show how techniques developed for one problem formulation can be adapted to solve a different formulation, thereby providing several novel adaptations to solve the different problem formulations. We also highlight the applicability of the techniques that handle discrete sequences to other related areas such as online anomaly detection and time series anomaly detection."
]
} |
1902.03155 | 2919816369 | In this paper, we introduce BINet, a neural network architecture for real-time multi-perspective anomaly detection in business process event logs. BINet is designed to handle both the control flow and the data perspective of a business process. Additionally, we propose a set of heuristics for setting the threshold of an anomaly detection algorithm automatically. We demonstrate that BINet can be used to detect anomalies in event logs not only on a case level but also on event attribute level. Finally, we demonstrate that a simple set of rules can be used to utilize the output of BINet for anomaly classification. We compare BINet to eight other state-of-the-art anomaly detection algorithms and evaluate their performance on an elaborate data corpus of 29 synthetic and 15 real-life event logs. BINet outperforms all other methods both on the synthetic as well as on the real-life datasets. | Probabilistic approaches estimate the probability distribution of the normal class and thus can detect anomalies as they come from a different distribution. An important probabilistic technique is the sliding window approach @cite_15 . In window-based anomaly detection, an anomaly score is assigned to each window in a sequence. Then the anomaly score of the sequence can be inferred by aggregating the window anomaly scores. Recently, Wressnegger et ,al. used this approach for intrusion detection and gave an elaborate evaluation in @cite_6 . While being inexpensive and easy to implement, sliding window approaches show a robust performance in finding anomalies in sequential data, especially within short regions @cite_17 . | {
"cite_N": [
"@cite_15",
"@cite_6",
"@cite_17"
],
"mid": [
"2129860818",
"2064274762",
"2038819732"
],
"abstract": [
"Intrusion detection systems rely on a wide variety of observable data to distinguish between legitimate and illegitimate activities. We study one such observable-sequences of system calls into the kernel of an operating system. Using system-call data sets generated by several different programs, we compare the ability of different data modeling methods to represent normal behavior accurately and to recognize intrusions. We compare the following methods: simple enumeration of observed sequences; comparison of relative frequencies of different sequences; a rule induction technique; and hidden Markov models (HMMs). We discuss the factors affecting the performance of each method and conclude that for this particular problem, weaker methods than HMMs are likely sufficient.",
"Detection methods based on n-gram models have been widely studied for the identification of attacks and malicious software. These methods usually build on one of two learning schemes: anomaly detection, where a model of normality is constructed from n-grams, or classification, where a discrimination between benign and malicious n-grams is learned. Although successful in many security domains, previous work falls short of explaining why a particular scheme is used and more importantly what renders one favorable over the other for a given type of data. In this paper we provide a close look on n-gram models for intrusion detection. We specifically study anomaly detection and classification using n-grams and develop criteria for data being used in one or the other scheme. Furthermore, we apply these criteria in the scope of web intrusion detection and empirically validate their effectiveness with different learning-based detection methods for client-side and service-side attacks.",
"This survey attempts to provide a comprehensive and structured overview of the existing research for the problem of detecting anomalies in discrete symbolic sequences. The objective is to provide a global understanding of the sequence anomaly detection problem and how existing techniques relate to each other. The key contribution of this survey is the classification of the existing research into three distinct categories, based on the problem formulation that they are trying to solve. These problem formulations are: 1) identifying anomalous sequences with respect to a database of normal sequences; 2) identifying an anomalous subsequence within a long sequence; and 3) identifying a pattern in a sequence whose frequency of occurrence is anomalous. We show how each of these problem formulations is characteristically distinct from each other and discuss their relevance in various application domains. We review techniques from many disparate and disconnected application domains that address each of these formulations. Within each problem formulation, we group techniques into categories based on the nature of the underlying algorithm. For each category, we provide a basic anomaly detection technique, and show how the existing techniques are variants of the basic technique. This approach shows how different techniques within a category are related or different from each other. Our categorization reveals new variants and combinations that have not been investigated before for anomaly detection. We also provide a discussion of relative strengths and weaknesses of different techniques. We show how techniques developed for one problem formulation can be adapted to solve a different formulation, thereby providing several novel adaptations to solve the different problem formulations. We also highlight the applicability of the techniques that handle discrete sequences to other related areas such as online anomaly detection and time series anomaly detection."
]
} |
1902.03155 | 2919816369 | In this paper, we introduce BINet, a neural network architecture for real-time multi-perspective anomaly detection in business process event logs. BINet is designed to handle both the control flow and the data perspective of a business process. Additionally, we propose a set of heuristics for setting the threshold of an anomaly detection algorithm automatically. We demonstrate that BINet can be used to detect anomalies in event logs not only on a case level but also on event attribute level. Finally, we demonstrate that a simple set of rules can be used to utilize the output of BINet for anomaly classification. We compare BINet to eight other state-of-the-art anomaly detection algorithms and evaluate their performance on an elaborate data corpus of 29 synthetic and 15 real-life event logs. BINet outperforms all other methods both on the synthetic as well as on the real-life datasets. | Distance-based novelty detection does not require a clean dataset, yet it is only partly applicable to process cases, as anomalous cases are usually very similar to normal ones. A popular distance-based approach is the one-class support vector machine (OC-SVM). Sch "olkopf et ,al. @cite_22 first used support vector machines @cite_1 for anomaly detection. | {
"cite_N": [
"@cite_1",
"@cite_22"
],
"mid": [
"2119821739",
"2105497548"
],
"abstract": [
"The support-vector network is a new learning machine for two-group classification problems. The machine conceptually implements the following idea: input vectors are non-linearly mapped to a very high-dimension feature space. In this feature space a linear decision surface is constructed. Special properties of the decision surface ensures high generalization ability of the learning machine. The idea behind the support-vector network was previously implemented for the restricted case where the training data can be separated without errors. We here extend this result to non-separable training data. High generalization ability of support-vector networks utilizing polynomial input transformations is demonstrated. We also compare the performance of the support-vector network to various classical learning algorithms that all took part in a benchmark study of Optical Character Recognition.",
"Suppose you are given some dataset drawn from an underlying probability distribution P and you want to estimate a \"simple\" subset S of input space such that the probability that a test point drawn from P lies outside of S equals some a priori specified ν between 0 and 1. We propose a method to approach this problem by trying to estimate a function f which is positive on S and negative on the complement. The functional form of f is given by a kernel expansion in terms of a potentially small subset of the training data; it is regularized by controlling the length of the weight vector in an associated feature space. We provide a theoretical analysis of the statistical performance of our algorithm. The algorithm is a natural extension of the support vector algorithm to the case of unlabelled data."
]
} |
1902.03155 | 2919816369 | In this paper, we introduce BINet, a neural network architecture for real-time multi-perspective anomaly detection in business process event logs. BINet is designed to handle both the control flow and the data perspective of a business process. Additionally, we propose a set of heuristics for setting the threshold of an anomaly detection algorithm automatically. We demonstrate that BINet can be used to detect anomalies in event logs not only on a case level but also on event attribute level. Finally, we demonstrate that a simple set of rules can be used to utilize the output of BINet for anomaly classification. We compare BINet to eight other state-of-the-art anomaly detection algorithms and evaluate their performance on an elaborate data corpus of 29 synthetic and 15 real-life event logs. BINet outperforms all other methods both on the synthetic as well as on the real-life datasets. | The core of BINet is a recurrent neural network, trained to predict the next event and its attributes. The architecture is influenced by the works of Evermann @cite_31 @cite_18 and Tax @cite_28 , who utilized long short-term memory @cite_8 (LSTM) networks for next event prediction, demonstrating their utility. LSTMs have been used for anomaly detection in different contexts like acoustic novelty detection @cite_5 and predictive maintenance @cite_21 . These applications mainly focus on the detection of anomalies in time series and not, like BINet, on multi-perspective anomaly detection in discrete sequences of events. | {
"cite_N": [
"@cite_18",
"@cite_8",
"@cite_28",
"@cite_21",
"@cite_5",
"@cite_31"
],
"mid": [
"",
"",
"2581522324",
"2474046725",
"1601124178",
"2610829762"
],
"abstract": [
"",
"",
"Predictive business process monitoring methods exploit logs of completed cases of a process in order to make predictions about running cases thereof. Existing methods in this space are tailor-made for specific prediction tasks. Moreover, their relative accuracy is highly sensitive to the dataset at hand, thus requiring users to engage in trial-and-error and tuning when applying them in a specific setting. This paper investigates Long Short-Term Memory (LSTM) neural networks as an approach to build consistently accurate models for a wide range of predictive process monitoring tasks. First, we show that LSTMs outperform existing techniques to predict the next event of a running case and its timestamp. Next, we show how to use models for predicting the next task in order to predict the full continuation of a running case. Finally, we apply the same approach to predict the remaining time, and show that this approach outperforms existing tailor-made methods.",
"Mechanical devices such as engines, vehicles, aircrafts, etc., are typically instrumented with numerous sensors to capture the behavior and health of the machine. However, there are often external factors or variables which are not captured by sensors leading to time-series which are inherently unpredictable. For instance, manual controls and or unmonitored environmental conditions or load may lead to inherently unpredictable time-series. Detecting anomalies in such scenarios becomes challenging using standard approaches based on mathematical models that rely on stationarity, or prediction models that utilize prediction errors to detect anomalies. We propose a Long Short Term Memory Networks based Encoder-Decoder scheme for Anomaly Detection (EncDec-AD) that learns to reconstruct 'normal' time-series behavior, and thereafter uses reconstruction error to detect anomalies. We experiment with three publicly available quasi predictable time-series datasets: power demand, space shuttle, and ECG, and two real-world engine datasets with both predictive and unpredictable behavior. We show that EncDec-AD is robust and can detect anomalies from predictable, unpredictable, periodic, aperiodic, and quasi-periodic time-series. Further, we show that EncDec-AD is able to detect anomalies from short time-series (length as small as 30) as well as long time-series (length as large as 500).",
"Acoustic novelty detection aims at identifying abnormal novel acoustic signals which differ from the reference normal data that the system was trained with. In this paper we present a novel unsupervised approach based on a denoising autoencoder. In our approach auditory spectral features are processed by a denoising autoencoder with bidirectional Long Short-Term Memory recurrent neural networks. We use the reconstruction error between the input and the output of the autoencoder as activation signal to detect novel events. The autoencoder is trained on a public database which contains recordings of typical in-home situations such as talking, watching television, playing and eating. The evaluation was performed on more than 260 different abnormal events. We compare results with state-of-theart methods and we conclude that our novel approach significantly outperforms existing methods by achieving up to 93.4 F-Measure.",
"Predicting the final state of a running process, the remaining time to completion or the next activity of a running process are important aspects of runtime process management. Runtime management requires the ability to identify processes that are at risk of not meeting certain criteria in order to offer case managers decision information for timely intervention. This in turn requires accurate prediction models for process outcomes and for the next process event, based on runtime information available at the prediction and decision point. In this paper, we describe an initial application of deep learning with recurrent neural networks to the problem of predicting the next process event. This is both a novel method in process prediction, which has previously relied on explicit process models in the form of Hidden Markov Models (HMM) or annotated transition systems, and also a novel application for deep learning methods."
]
} |
1902.03192 | 2950973314 | Recently, the field of deep learning has received great attention by the scientific community and it is used to provide improved solutions to many computer vision problems. Convolutional neural networks (CNNs) have been successfully used to attack problems such as object recognition, object detection, semantic segmentation, and scene understanding. The rapid development of deep learning goes hand by hand with the adaptation of GPUs for accelerating its processes, such as network training and inference. Even though FPGA design exists long before the use of GPUs for accelerating computations and despite the fact that high-level synthesis (HLS) tools are getting more attractive, the adaptation of FPGAs for deep learning research and application development is poor due to the requirement of hardware design related expertise. This work presents a workflow for deep learning mobile application acceleration on small low-cost low-power FPGA devices using HLS tools. This workflow eases the design of an improved version of the SqueezeJet accelerator used for the speedup of mobile-friendly low-parameter ImageNet class CNNs, such as the SqueezeNet v1.1 and the ZynqNet. Additionally, the workflow includes the development of an HLS-driven analytical model which is used for performance estimation of the accelerator. This model can be also used to direct the design process and lead to future design improvements and optimizations. | ZynqNet describes a CNN architecture and an HLS design for the acceleration of this network. This approach shows that for achieving the desirable results, the problem must be transformed to aim the processing on the selected platform. ZynqNet derived from SqueezeNet by replacing the combination of convolutional and maxpool layers with a convolutional layer having increased stride @cite_11 . This transformation simplifies the accelerator design; by implementing a convolutional layer and a global pooling layer, the ZynqNet accelerator can process the whole CNN except the last softmax layer. Convolutional layer acceleration is achieved by calculating multiple output feature-map channels in parallel using processing elements (PEs) which fully unroll the calculation of a @math kernel. | {
"cite_N": [
"@cite_11"
],
"mid": [
"2123045220"
],
"abstract": [
"Most modern convolutional neural networks (CNNs) used for object recognition are built using the same principles: Alternating convolution and max-pooling layers followed by a small number of fully connected layers. We re-evaluate the state of the art for object recognition from small images with convolutional networks, questioning the necessity of different components in the pipeline. We find that max-pooling can simply be replaced by a convolutional layer with increased stride without loss in accuracy on several image recognition benchmarks. Following this finding -- and building on other recent work for finding simple network structures -- we propose a new architecture that consists solely of convolutional layers and yields competitive or state of the art performance on several object recognition datasets (CIFAR-10, CIFAR-100, ImageNet). To analyze the network we introduce a new variant of the \"deconvolution approach\" for visualizing features learned by CNNs, which can be applied to a broader range of network structures than existing approaches."
]
} |
1902.03192 | 2950973314 | Recently, the field of deep learning has received great attention by the scientific community and it is used to provide improved solutions to many computer vision problems. Convolutional neural networks (CNNs) have been successfully used to attack problems such as object recognition, object detection, semantic segmentation, and scene understanding. The rapid development of deep learning goes hand by hand with the adaptation of GPUs for accelerating its processes, such as network training and inference. Even though FPGA design exists long before the use of GPUs for accelerating computations and despite the fact that high-level synthesis (HLS) tools are getting more attractive, the adaptation of FPGAs for deep learning research and application development is poor due to the requirement of hardware design related expertise. This work presents a workflow for deep learning mobile application acceleration on small low-cost low-power FPGA devices using HLS tools. This workflow eases the design of an improved version of the SqueezeJet accelerator used for the speedup of mobile-friendly low-parameter ImageNet class CNNs, such as the SqueezeNet v1.1 and the ZynqNet. Additionally, the workflow includes the development of an HLS-driven analytical model which is used for performance estimation of the accelerator. This model can be also used to direct the design process and lead to future design improvements and optimizations. | In Angel-Eye @cite_7 , a design flow for mapping CNNs onto embedded FPGA devices is proposed. This design flow includes a dynamic fixed-point quantization strategy, a software controlled hardware architecture with @math convolution kernel support, and a run-time workflow which allows a single frame to be processed by multiple CNNs. Since real-time processing is of main concern, Angel-Eye uses a batch size of one in order to minimize latency. | {
"cite_N": [
"@cite_7"
],
"mid": [
"2616014673"
],
"abstract": [
"Convolutional neural network (CNN) has become a successful algorithm in the region of artificial intelligence and a strong candidate for many computer vision algorithms. But the computation complexity of CNN is much higher than traditional algorithms. With the help of GPU acceleration, CNN-based applications are widely deployed in servers. However, for embedded platforms, CNN-based solutions are still too complex to be applied. Various dedicated hardware designs on field-programmable gate arrays (FPGAs) have been carried out to accelerate CNNs, while few of them explore the whole design flow for both fast deployment and high power efficiency. In this paper, we investigate state-of-the-art CNN models and CNN-based applications. Requirements on memory, computation and the flexibility of the system are summarized for mapping CNN on embedded FPGAs. Based on these requirements, we propose Angel-Eye, a programmable and flexible CNN accelerator architecture, together with data quantization strategy and compilation tool. Data quantization strategy helps reduce the bit-width down to 8-bit with negligible accuracy loss. The compilation tool maps a certain CNN model efficiently onto hardware. Evaluated on Zynq XC7Z045 platform, Angel-Eye is @math faster and @math better in power efficiency than peer FPGA implementation on the same platform. Applications of VGG network, pedestrian detection and face alignment are used to evaluate our design on Zynq XC7Z020. NIVIDA TK1 and TX1 platforms are used for comparison. Angel-Eye achieves similar performance and delivers up to @math better energy efficiency."
]
} |
1902.03192 | 2950973314 | Recently, the field of deep learning has received great attention by the scientific community and it is used to provide improved solutions to many computer vision problems. Convolutional neural networks (CNNs) have been successfully used to attack problems such as object recognition, object detection, semantic segmentation, and scene understanding. The rapid development of deep learning goes hand by hand with the adaptation of GPUs for accelerating its processes, such as network training and inference. Even though FPGA design exists long before the use of GPUs for accelerating computations and despite the fact that high-level synthesis (HLS) tools are getting more attractive, the adaptation of FPGAs for deep learning research and application development is poor due to the requirement of hardware design related expertise. This work presents a workflow for deep learning mobile application acceleration on small low-cost low-power FPGA devices using HLS tools. This workflow eases the design of an improved version of the SqueezeJet accelerator used for the speedup of mobile-friendly low-parameter ImageNet class CNNs, such as the SqueezeNet v1.1 and the ZynqNet. Additionally, the workflow includes the development of an HLS-driven analytical model which is used for performance estimation of the accelerator. This model can be also used to direct the design process and lead to future design improvements and optimizations. | In @cite_12 a latency-driven design method is presented as an extension of the fpgaConvNet modeling framework @cite_10 . This work models CNNs using the synchronous dataflow (SDF) model of computation. CNNs are interpreted as directed acyclic graphs (DAGs) whose nodes are mapped to hardware building blocks which are interconnected to form the final SDF graph. The SDF model of computation allows the generation of static schedules of execution and the calculation of the amount of buffer memory between the interconnected hardware building blocks. The SDF graph is partitioned along its depth and a single flexible reference architecture is generated which enables the execution of all the subgraphs. In contrast to Angle-Eye, this reference architecture is tailored to a specific CNN and it is optimized in terms of latency. Their framework produces synthesizable Vivado HLS code for the resulting architecture. | {
"cite_N": [
"@cite_10",
"@cite_12"
],
"mid": [
"2513568085",
"2761262947"
],
"abstract": [
"Convolutional Neural Networks (ConvNets) are a powerful Deep Learning model, providing state-of-the-art accuracy to many emerging classification problems. However, ConvNet classification is a computationally heavy task, suffering from rapid complexity scaling. This paper presents fpgaConvNet, a novel domain-specific modelling framework together with an automated design methodology for the mapping of ConvNets onto reconfigurable FPGA-based platforms. By interpreting ConvNet classification as a streaming application, the proposed framework employs the Synchronous Dataflow (SDF) model of computation as its basis and proposes a set of transformations on the SDF graph that explore the performance-resource design space, while taking into account platform-specific resource constraints. A comparison with existing ConvNet FPGA works shows that the proposed fully-automated methodology yields hardware designs that improve the performance density by up to 1.62x and reach up to 90.75 of the raw performance of architectures that are hand-tuned for particular ConvNets.",
"In recent years, Convolutional Neural Networks (ConvNets) have become the quintessential component of several state-of-the-art Artificial Intelligence tasks. Across the spectrum of applications, the performance needs vary significantly, from high-throughput image recognition to the very low-latency requirements of autonomous cars. In this context, FPGAs can provide a potential platform that can be optimally configured based on different performance requirements. However, with the increasing complexity of ConvNet models, the architectural design space becomes overwhelmingly large, asking for principled design flows that address the application-level needs. This paper presents a latency-driven design methodology for mapping ConvNets on FPGAs. The proposed design flow employs novel transformations over a Synchronous Dataflow-based modelling framework together with a latency-centric optimisation procedure in order to efficiently explore the design space targeting low-latency designs. Quantitative evaluation shows large improvements in latency when latency-driven optimisation is in place yielding designs that improve the latency of AlexNet by 73.54× and VGG16 by 5.61× over throughput-optimised designs."
]
} |
1902.03185 | 2913294020 | The human ability to coordinate and cooperate has been vital to the development of societies for thousands of years. While it is not fully clear how this behavior arises, social norms are thought to be a key factor in this development. In contrast to laws set by authorities, norms tend to evolve in a bottom-up manner from interactions between members of a society. While much behavior can be explained through the use of social norms, it is difficult to measure the extent to which they shape society as well as how they are affected by other societ al dynamics. In this paper, we discuss the design and evaluation of a reinforcement learning model for understanding how the opportunity to choose who you interact with in a society affects the overall societ al outcome and the strength of social norms. We first study the emergence of norms and then the emergence of cooperation in presence of norms. In our model, agents interact with other agents in a society in the form of repeated matrix-games: coordination games and cooperation games. In particular, in our model, at each each stage, agents are either able to choose a partner to interact with or are forced to interact at random and learn using policy gradients. | While researchers can employ learning strategies in agents to simulate the emergence of norms, it is a difficult task to build simulations that encompass the natural physical and social constraints that individuals are faced within a society. Nevertheless, investigations into normative behavior have been revealing. proposed a replicable multi-agent learning framework for social norm emergence in which agents interact at random in a society @cite_16 . In this simulation, agents had access to actions taken by the opponent but not their identity. They demonstrated that agents trained with a hill-climbing technique, Win-or-Lose-Fast (WoLF) @cite_11 were capable of converging to a norm in a driving simulation that required all agents in a population to drive either on the left side or the right side of the road for maximum societ al gain @cite_16 . In another set of experiments, test the outcome of this framework while constraining agents to interactions with their neighbors @cite_23 . This has since been further extended to agent societies considering a variety of network topologies @cite_4 that have proven suitable for the emergence of norms and conventions as they provide an effective infrastructure in which interactions within a social network can take place. | {
"cite_N": [
"@cite_16",
"@cite_4",
"@cite_23",
"@cite_11"
],
"mid": [
"92376239",
"2172193467",
"1565938022",
"2120327309"
],
"abstract": [
"Behavioral norms are key ingredients that allow agent coordination where societ al laws do not sufficiently constrain agent behaviors. Whereas social laws need to be enforced in a top-down manner, norms evolve in a bottom-up manner and are typically more self-enforcing. While effective norms can significantly enhance performance of individual agents and agent societies, there has been little work in multiagent systems on the formation of social norms. We propose a model that supports the emergence of social norms via learning from interaction experiences. In our model, individual agents repeatedly interact with other agents in the society over instances of a given scenario. Each interaction is framed as a stage game. An agent learns its policy to play the game over repeated interactions with multiple agents. We term this mode of learning social learning, which is distinct from an agent learning from repeated interactions against the same player. We are particularly interested in situations where multiple action combinations yield the same optimal payoff. The key research question is to find out if the entire population learns to converge to a consistent norm. In addition to studying such emergence of social norms among homogeneous learners via social learning, we study the effects of heterogeneous learners, population size, multiple social groups, etc.",
"Social norms play a pivotal role in sustaining social order by regulating individual behaviors in a society. In normative multiagent systems, social norms have been used as an efficient mechanism to govern virtual agent societies towards cooperation and coordination. In this paper, we study the emergence of social norms via learning from repeated local interactions in networked agent societies. We propose a collective learning framework, which imitates the opinion aggregation process in human decision making, to study the impact of agent local collective behaviors on norm emergence in different situations. In the framework, each agent interacts repeatedly with all of its neighbors. At each step, an agent first takes a best-response action towards each of its neighbors and then combines all of these actions into a final action using ensemble learning methods. We conduct extensive experiments to evaluate the framework with respect to different network topologies, learning strategies, numbers of actions, and so on. Experimental results reveal some significant insights into norm emergence in networked agent societies achieved through local collective behaviors.",
"A social norm is a behavior that emerges as a convention within society without any direction from a central authority. Social norms emerge as repeated interactions between individuals give rise to biases toward actions or behaviors which spread through the society until one behavior is adapted as the default behavior, even when multiple acceptable behaviors exist. Of particular interest to us is how and when norms emerge in social networks, which provide a framework for individuals to interact routinely. We study how quickly norms converge in social networks depending on parameters such as the topology of the network, population size, neighborhood size, and number of behavior alternatives. Our research can be used to model and analyze popular social networks on the Internet such as Facebook, Flickr, and Digg. In addition, it can be used to predict how norms emerge and spread in human societies, ranging from routine decisions like which side of the road to drive on to social trends such as the green phenomenon.",
"Abstract Learning to act in a multiagent environment is a difficult problem since the normal definition of an optimal policy no longer applies. The optimal policy at any moment depends on the policies of the other agents. This creates a situation of learning a moving target. Previous learning algorithms have one of two shortcomings depending on their approach. They either converge to a policy that may not be optimal against the specific opponents' policies, or they may not converge at all. In this article we examine this learning problem in the framework of stochastic games. We look at a number of previous learning algorithms showing how they fail at one of the above criteria. We then contribute a new reinforcement learning technique using a variable learning rate to overcome these shortcomings. Specifically, we introduce the WoLF principle, “Win or Learn Fast”, for varying the learning rate. We examine this technique theoretically, proving convergence in self-play on a restricted class of iterated matrix games. We also present empirical results on a variety of more general stochastic games, in situations of self-play and otherwise, demonstrating the wide applicability of this method."
]
} |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.