paper_name stringlengths 11 170 | text stringlengths 8.07k 307k | summary stringlengths 152 6.16k | paper_id stringlengths 43 43 |
|---|---|---|---|
Property Controllable Variational Autoencoder via Invertible Mutual Dependence | 1 INTRODUCTION . Important progress has been made towards learning the underlying low-dimensional representation and generative process of complex high dimensional data such as images ( Pu et al. , 2016 ) , natural languages ( Bowman et al. , 2016 ) , chemical molecules ( Kadurin et al. , 2017 ; Guo et al. , 2019 ) and... | To encourage disentanglement in the latent space of a variational autoencoder (VAE), the authors propose to learn two sets of latent z and w: the dimensions of w are independent of each other and each dimension w_i maps to a known ground truth generating factor y_i. Latent z captures all the other factors. The well stu... | SP:71d80c38253a4fd86d8e076b2dee8d0c47da4911 |
OPAL: Offline Primitive Discovery for Accelerating Offline Reinforcement Learning | 1 INTRODUCTION . Reinforcement Learning ( RL ) systems have achieved impressive performance in a variety of online settings such as games ( Silver et al. , 2016 ; Tesauro , 1995 ; Brown & Sandholm , 2019 ) and robotics ( Levine et al. , 2016 ; Dasari et al. , 2019 ; Peters et al. , 2010 ; Parmas et al. , 2019 ; Pinto &... | In the RL setting, this paper tackles the case where an agent may have access to large amounts of offline experience data. The objective of the work is to find an effective way to leverage this data for finding temporally extended primitive behaviors. The paper provides results that show how performing offline primitiv... | SP:b2e0fd72f2a599a9cb288618decfcd709d712f38 |
Federated Learning of a Mixture of Global and Local Models | 1 INTRODUCTION . With the proliferation of mobile phones , wearable devices , tablets , and smart home devices comes an increase in the volume of data captured and stored on them . This data contains a wealth of potentially useful information to the owners of these devices , and more so if appropriate machine learning ... | of the paper: The paper proposes a new formulation for the federated learning problem, in which each agent has its local model, and a penalty term is added to the objective function to control the deviation of these local models from their average. Next, the authors develop a randomized algorithm to tackle this problem... | SP:f2ba6d73cecdf611e6f58c93fb88e9f3dbe3bb24 |
Neuron Activation Analysis for Multi-Joint Robot Reinforcement Learning | 1 INTRODUCTION . Convolutional Neural Networks ( CNN ) are well known to demonstrate a strong general feature extraction capability in lower network layers . In these networks feature kernels can not only be visualized , pre-trained general feature extractors can also be reused for efficient network learning . Recent e... | The paper presents a technique to compare networks trained to solve similar tasks trained in different context. The considered task is reaching with a robotic planar arm; the considered context is varied varying the robot degrees of freedom. The goal of the paper is to find correlations across neural activity patterns ... | SP:df2c981f4cfc3734e8f42c4e84368e90d931529b |
Adaptive Personalized Federated Learning | 1 INTRODUCTION . With the massive amount of data generated by the proliferation of mobile devices and the internet of things ( IoT ) , coupled with concerns over sharing private information , collaborative machine learning and the use of federated optimization ( FO ) is often crucial for the deployment of large-scale m... | This paper studies the personalization aspect of the federated learning problem. The authors propose a new framework in which they replace the common global model in the original federated learning formulation with a convex combination of the global model and a local model. They later introduce an adaptive optimization... | SP:bdb67d79c8c71ff8761649620a110bd8ff2353fe |
Topology-Aware Segmentation Using Discrete Morse Theory | 1 INTRODUCTION . Segmenting objects while preserving their global structure is a challenging yet important problem . Various methods have been proposed to encourage neural networks to preserve fine details of objects ( Long et al. , 2015 ; He et al. , 2017 ; Chen et al. , 2014 ; 2018 ; 2017 ) . Despite their high per-p... | The paper focus on the segmentation of images in which the correct topology of the segmentation an important role plays. The key idea of the paper is the use of Discrete Morse Theory to identify those areas of the segmentation/likelihood map that are important for ensuring the correct topology of the segmentation. This... | SP:8c67767eb8c691157f0807c4c8393749b78bdf47 |
Decoupling Representation Learning from Reinforcement Learning | In an effort to overcome limitations of reward-driven feature learning in deep reinforcement learning ( RL ) from images , we propose decoupling representation learning from policy learning . To this end , we introduce a new unsupervised learning ( UL ) task , called Augmented Temporal Contrast ( ATC ) , which trains a... | This paper presents a new unsupervised learning method for learning latent representations for visual RL control domains. The method, Augmented Temporal Contrast (ATC), can be used alone to learn a representation to be combined with an RL algorithm, or as an auxiliary task in an end-to-end system. ATC matches or outper... | SP:9c3d7291bb936e41d94e0357e2085cb1621d4f3a |
TopoTER: Unsupervised Learning of Topology Transformation Equivariant Representations | 1 INTRODUCTION . Graphs provide a natural and efficient representation for non-Euclidean data , such as brain networks , social networks , citation networks , and 3D point clouds . Graph Convolutional Neural Networks ( GCNNs ) ( Bronstein et al. , 2017 ) have been proposed to generalize the CNNs to learn representation... | This paper develops a framework for unsupervised learning of graphs. The goal is to build graph representation using an encoder that is useful for downstream tasks such as graph classification. The representation is computed with an encoder $E$ applied to a graph data $(X,A)$, containing vertex data $X$ and adjacency m... | SP:9a92f7adeccc0f66836e3ddfb6bd5af67bdf77e4 |
gradSim: Differentiable simulation for system identification and visuomotor control | 1 INTRODUCTION . Accurately predicting the dynamics and physical characteristics of objects from image sequences is a long-standing challenge in computer vision . This end-to-end reasoning task requires a fundamental understanding of both the underlying scene dynamics and the imaging process . Imagine watching a short ... | This paper presents a framework for performing both differentiable physics simulations and differentiable rendering. This fully differentiable simulation and rendering pipeline is then employed to perform system identification tasks, directly from video frames, being able to match or outperform both visual-based and st... | SP:ce42fb1c2d8241aeb60250b7a0229411a2dcfa81 |
Proximal Gradient Descent-Ascent: Variable Convergence under KŁ Geometry | 1 INTRODUCTION . Minimax optimization is a classical optimization framework that has been widely applied in various modern machine learning applications , including game theory Ferreira et al . ( 2012 ) , generative adversarial networks ( GANs ) Goodfellow et al . ( 2014 ) , adversarial training Sinha et al . ( 2017 ) ... | In this paper, the authors analyze the convergence of a proximal gradient descent ascent (GDA) method when applied to non-convex strongly concave functions. To establish convergence results, the authors show that proximal-GDA admits a novel Lyapunov function that monotonically decreases at every iteration. Along with K... | SP:f958cd0237ec397729161f29cab903af40716fd3 |
Learning advanced mathematical computations from examples | 1 Introduction . Scientists solve problems of mathematics by applying rules and computational methods to the data at hand . These rules are derived from theory , they are taught in schools or implemented in software libraries , and guarantee that a correct solution will be found . Over time , mathematicians have develo... | This paper shows that transformer models can be used to accurately learn advanced mathematical computations from millions of examples. The problems are drawn from the fields of differential equations and control theory. The selected problems are ones that are solvable using known algorithms; however, these algorithms... | SP:79b19c6490c2ea5cab56666927520888191a83a7 |
Neural Random Projection: From the Initial Task To the Input Similarity Problem | 1 INTRODUCTION . Evaluating object similarity is an important area in machine learning literature . It is used in various applications such as search query matching , image similarity search , recommender systems , clustering , classification . In practice , the quality of similarity evaluation methods depends on the d... | The paper studies the usage of the representations developed in the last layer of a neural network as a way to measure the similarity between input patterns. The fundamental idea revolves around the concept of orthogonal weight matrices, to decorrelate the activations of the neurons, and which would definitely enrich t... | SP:504c896a68c2e0154232d2f2214e8a499d941b60 |
Learning to Actively Learn: A Robust Approach | This work proposes a procedure for designing algorithms for specific adaptive data collection tasks like active learning and pure-exploration multi-armed bandits . Unlike the design of traditional adaptive algorithms that rely on concentration of measure and careful analysis to justify the correctness and sample comple... | The paper "Learning to Actively Learn" proposes a differentiable procedure to design algorithms for adaptive data collection tasks. The framework is based on the idea of making use of a measure of problem convexity (each problem is parametrized by a parameter theta) to solve solving a min-max objective over policies. T... | SP:b5366ec9f9cf6d872098a4f610801990b35b29d7 |
A Technical and Normative Investigation of Social Bias Amplification | 1 INTRODUCTION . The machine learning community is becoming increasingly cognizant of problems surrounding fairness and bias , and correspondingly a plethora of new algorithms and metrics are being proposed ( see e.g. , Mehrabi et al . ( 2019 ) for a review ) . The gatekeepers checking the systems to be deployed often ... | The paper builds on the "bias amplification" aspect of fairness in machine learning literature i.e. the tendency of models to make predictions that are biased in a way that they amplify societal correlations. The paper claims three major contributions: a metric, discussion about the dependence of bias measurements on r... | SP:53254b85dc51beed47567630a243e95537bb65ed |
Dynamic of Stochastic Gradient Descent with State-dependent Noise | 1 INTRODUCTION . Deep learning has achieved great success in various AI applications , such as computer vision , natural language processing , and speech recognition ( He et al. , 2016b ; Vaswani et al. , 2017 ; He et al. , 2016a ) . Stochastic gradient descent ( SGD ) and its variants are the mainstream methods to tra... | This paper proposes an analysis for an approximate dynamic of SGD which captures the heavy-tailed noise distributions seen practically at local minima. The authors derive this new dynamic (which they call Power-law dynamic) using basic principles and the assumption that the noise variance depends on the state. The dyna... | SP:016ad5eea3a81bb32ae907535e694d8171f996a5 |
On Linear Identifiability of Learned Representations | 1 INTRODUCTION . An increasingly common methodology in machine learning is to improve performance on a primary down-stream task by first learning a high-dimensional representation of the data on a related , proxy task . In this paradigm , training a model reduces to fine-tuning the learned representations for optimal p... | In this paper, the authors address the model identifiability in a general setting that can be adapted to several recent deep learning models (DNN supervised learning, CPC, BERT and GPT). Since model parameters (NN weights) are not identifiable, the authors hypothesize that vector f and g can be identifiably up to a lin... | SP:45ebf4f0cb747eadb32b5254b75d142755f67af6 |
Deep Quotient Manifold Modeling | 1 INTRODUCTION . Real-world data are usually considered to involve a multi-manifold structure by having discrete features as well as continuous features ; continuous features such as size or location induce a smooth manifold structure in general , whereas discrete features such as digit-class or a new object in the bac... | The author extends generative models with multi-generators by restricting the generators to share weights and all bias to be regularized in order to enforce that the inverse maps of the generators can be represented by a single encoder. The regularizer proposed minimizes an upper bound the the sum of the bias variances... | SP:a3b2fa4479b45b1a59e573b452c73cae507485ba |
Crowd-sourced Phrase-Based Tokenization for Low-Resourced Neural Machine Translation: The case of Fon Language | 1 INTRODUCTION . We would like to start by sharing with you this Fon sentence : « m¢tà m¢tà w¢ zìnwó h¢n wa aligbo m¢ » . How would you tokenize this ? What happens if we implement the standard method of splitting the sentence into its word elements ( either using the space delimiter or using subword units ) ? m¢tà m¢t... | The authors investigate different tokenization methods for the translation between French and Fon (an African low-resource language). This means that they compare different ways to construct the input and output vocabularies of a neural machine translation (NMT) system. They further propose their own way to create thos... | SP:021a44760567c1cf821f9388b6812a24aa708755 |
Federated Continual Learning with Weighted Inter-client Transfer | 1 INTRODUCTION . Continual learning ( Thrun , 1995 ; Kumar & Daume III , 2012 ; Ruvolo & Eaton , 2013 ; Kirkpatrick et al. , 2017 ; Schwarz et al. , 2018 ) describes a learning scenario where a model continuously trains on a sequence of tasks ; it is inspired by the human learning process , as a person learns to perfor... | This paper investigates a new problem – federated continual learning by Federated Weighted Inter-client Transfer. The key idea is to decompose the network weights into global federated parameters and sparse task-specific parameters such that each client can selectively receive knowledge from other clients by taking a w... | SP:dc569d8f517e19b940c774679a98c14eb8272919 |
Sparse Quantized Spectral Clustering | 1 INTRODUCTION . Sparsifying , quantizing , and/or performing other entry-wise nonlinear operations on large matrices can have many benefits . Historically , this has been used to develop iterative algorithms for core numerical linear algebra problems ( Achlioptas & McSherry , 2007 ; Drineas & Zouzias , 2011 ) . More r... | This work considers the effect of sparsification, quantization and non-linear transormations on the spectrum of a random matrix with respect to performance in downstream applications like clustering. Eigen decomposition of large matrices is computationally very expensive and therefore methods like sparsification is use... | SP:517da79468af9d50729358716877de22c41431a9 |
Recurrent Independent Mechanisms | Learning modular structures which reflect the dynamics of the environment can lead to better generalization and robustness to changes which only affect a few of the underlying causes . We propose Recurrent Independent Mechanisms ( RIMs ) , a new recurrent architecture in which multiple groups of recurrent cells operate... | The authors argue that the world consists of largely independent causal mechanisms that sparsely interact. The authors propose a new kind of recurrent network (RIM) that presumably distills this world view into inductive biases. RIMs consist of largely independent recurrent modules that are sparsely activated and inter... | SP:351283f0a33b5d1eb8d54d4fb74e93f0505b051c |
Deep Reinforcement Learning with Causality-based Intrinsic Reward | 1 INTRODUCTION . Reinforcement learning ( RL ) is a powerful approach towards dealing with sequential decisionmaking problems . Combined with deep neural networks , deep reinforcement learning ( DRL ) has been applied in a variety of fields such as playing video games ( Mnih et al. , 2015 ; Vinyals et al. , 2019 ; Bern... | The paper proposes a deep reinforcement learning algorithm of advantage actor-critic (A2C), which firstly learns the causal structure of the environment and then leverages the learned causal information to assist policy learning. The causal structure is computed by calculating Average Causal Effect (ACE) between differe... | SP:6283566c3a3868af63b1721a727aad52eab1aec8 |
Adaptive Optimizers with Sparse Group Lasso | 1 INTRODUCTION . With the development of deep learning , deep neural network ( DNN ) models have been widely used in various machine learning scenarios such as search , recommendation and advertisement , and achieved significant improvements . In the last decades , different kinds of optimization methods based on the v... | This work studies the adaptive proximal gradient descent method, and specifically studies the group sparsity. To encourage the group sparsity, a regularizer which is a combination of $\ell_1$ norm, block $\ell_1$ norm and $\ell_2$ norm square is used. This paper gives the update rule of the proximal gradient with the s... | SP:d98564657b7cd55efa243520c5bd7ef8be405a26 |
Rethinking Architecture Selection in Differentiable NAS | 1 INTRODUCTION . Neural Architecture Search ( NAS ) has been drawing increasing attention in both academia and industry for its potential to automatize the process of discovering high-performance architectures , which have long been handcrafted . Early works on NAS deploy Evolutionary Algorithm ( Stanley & Miikkulainen... | In one-shot differentiable NAS, a supergraph is usually trained (via bilevel optimization as in DARTS, or other approximations to bilevel such as gumbel softmax, etc). After supergraph training, a final architecture is obtained by taking the operator at each edge which has the highest architecture weight magnitude. Thi... | SP:eceef2daaa86f86534b3b33ca96c19f0b52e20b7 |
On Episodes, Prototypical Networks, and Few-Shot Learning | 1 Introduction . The problem of few-shot learning ( FSL ) – classifying examples from previously unseen classes given only a handful of training data – has considerably grown in popularity within the machine learning community in the last few years . The reason is likely twofold . First , being able to perform well on ... | The paper's starting point is the question whether the episodic training is beneficial, or not, for FSL / Prototypical Networks. The work can be seen as a follow-up of the recent works showing that simple baselines can outperform rather sophisticated few-shot learning models. Towards answering this question, this paper... | SP:173177f78449ef09647670389b0ffba1e35db0ba |
Practical Real Time Recurrent Learning with a Sparse Approximation | 1 Introduction . Recurrent neural networks ( RNNs ) have been successfully applied to a wide range of sequence learning tasks , including text-to-speech ( Kalchbrenner et al. , 2018 ) , language modeling ( Dai et al. , 2019 ) , automatic speech recognition ( Amodei et al. , 2016 ) , translation ( Chen et al. , 2018 ) a... | This work presents a method, named SnAp, that takes Real-time Recurrent Learning derivations and proposes to approximate its computations with sparse approximations to make them more computational tractable. The method is an alternative to overcome the truncation in backpropagation through time (BPTT) over long term te... | SP:929103e44d7ee2bc3b8a4d0c9e523c082a17fb44 |
Degree-Quant: Quantization-Aware Training for Graph Neural Networks | 1 INTRODUCTION . GNNs have received substantial attention in recent years due to their ability to model irregularly structured data . As a result , they are extensively used for applications as diverse as molecular interactions ( Duvenaud et al. , 2015 ; Wu et al. , 2017 ) , social networks ( Hamilton et al. , 2017 ) ,... | The authors propose a new technique for quantization aware training of neural networks that is specially suited for graph neural networks. They do a good job of motivating the problem by demonstrating that the large variation of input degree in GNNs can lead to unique challenges for numerical precision, forcing a compr... | SP:e5152f19fbd60b76c867e34096a7ba19b2ed6af4 |
Cluster-Former: Clustering-based Sparse Transformer for Question Answering | 1 INTRODUCTION . Long-range contextual understanding has proven critical in many natural language processing ( NLP ) tasks . For example , the relevant context for correctly answering an open-domain question can arch over thousands of words . Encoding long sequences via deep neural networks , however , has remained an ... | Cluster-former is the latest proposal for enabling transformers to deal with long input sequences. Such sequences are particularly problematic for problems like question answering, QA, (or summarization), where the context can be arbitrarily long, and effectively open-ended when the setup includes a context retrieval c... | SP:031cbb9fd369d00fa867901cf650c777d356d853 |
Powers of layers for image-to-image translation | 1 INTRODUCTION . Neural networks define arbitrarily complex functions involved in discriminative or generative tasks by stacking layers , as supported by the universal approximation theorem ( Hornik et al. , 1989 ; Montúfar , 2014 ) . More precisely , the theorem states that stacking a number of basic blocks can approx... | This paper proposes an unpaired image-to-image translation method which applies a pre-trained auto-encoder and a latent feature transformer (single block) to perform iterative image transformation. A progressive training and warm-up strategy is used to settle the numerical exponentiation effects caused by powers of lay... | SP:31e67caf860b47d871f17848355cd91b65830f59 |
Partial Rejection Control for Robust Variational Inference in Sequential Latent Variable Models | 1 INTRODUCTION . Exact inference in latent variable models is usually intractable . Markov Chain Monte-Carlo ( MCMC ) ( Andrieu et al. , 2003 ) and variational inference ( VI ) methods ( Blei et al. , 2017 ) , are commonly employed in such models to make inference tractable . While MCMC has been the traditional method ... | This paper describes an SMC algorithm to sample the posterior distribution of latent states $p_\theta(z_{1:T}|x_{1:T})$ in a latent variable models $p_\theta(x_{1:T}, z_{1:T})$. The authors consider a completely general setting (the authors assume Eq.(1) but clearly there is nothing to assume here, this the standard Ba... | SP:26f9757d1a510fe264bda0570cea50301383c39a |
Flowtron: an Autoregressive Flow-based Generative Network for Text-to-Speech Synthesis | 1 INTRODUCTION . Current speech synthesis methods do not give the user enough control over how speech actually sounds . Automatically converting text to audio that successfully communicates the text was achieved a long time ago ( Umeda et al. , 1968 ; Badham et al. , 1983 ) . However , communicating only the text infor... | This paper presents a text-to-speech synthesis system, called Flowtron which uses a normalizing flow to generate a sequence of mel-spectrogram frames. The difference between the proposed Flowtron and the previously prosed flow-based methods is, the authors argue as the main contributions, its ability to produce more di... | SP:798bd74ff3a9e5b08eba7c3d90b5c85494cb48a8 |
Are Neural Rankers still Outperformed by Gradient Boosted Decision Trees? | Despite the success of neural models on many major machine learning problems , their effectiveness on traditional Learning-to-Rank ( LTR ) problems is still not widely acknowledged . We first validate this concern by showing that most recent neural LTR models are , by a large margin , inferior to the best publicly avai... | - The paper argues that neural models perform significantly worse than GBDT models on some learning to rank benchmarks. It first conducts a set of experiments to show that GBDT outperforms some neural rankers. Then, it presents a few tweaks related to feature transformation and data augmentation to improve the performa... | SP:a97b9cd6237040dc602bed2c66af26143847c37f |
Neural spatio-temporal reasoning with object-centric self-supervised learning | 1 INTRODUCTION . Artificial intelligence research has long been divided into rule-based approaches and statistical models . Neural networks , a classic example of the statistical approach , certainly have limitations despite their massive popularity and success . For example , experiments with two recently released vid... | the paper propose to tackle visual reasoning problem in videos. The proposed solution is to combine MONET (Burgess et al., 2019) with self-attention mechanism (Vaswani et al., 2017) to first encode images into object-centric encodings and aggregate the encodings using self-attention to make the final prediction. The me... | SP:e56c1cfe3a5303c1176c9778ef1ea75855d7e20f |
Influence Estimation for Generative Adversarial Networks | 1 INTRODUCTION . Generative adversarial networks ( GANs ) proposed by Goodfellow et al . ( 2014 ) are a powerful subclass of generative model , which is successfully applied to a number of image generation tasks ( Antoniou et al. , 2017 ; Ledig et al. , 2017 ; Wu et al. , 2016 ) . The expansion of the applications of G... | The paper presents an influence estimation method for GANs. It discusses why previous approaches on influence estimation cannot be easily extended to GANs. It proposes to use Jacobian of the gradient of discriminator’s loss with respect to the generator’s parameters to learn how absence of an instance in the discrimina... | SP:1d3fbd26ee829b120b08d1d474743606d3f72292 |
A Design Space Study for LISTA and Beyond | 1 INTRODUCTION . The signal processing and optimization realm has an everlasting research enthusiasm on addressing ill-conditioned inverse problems , that are often regularized by handcrafted model-based priors , such as sparse coding , low-rank matrix fitting and conditional random fields . Since closed-form solutions... | This paper studies a very interesting new problem of assessing unrolled models in a broader context using NAS methods. LISTA-style unrolling has been popular for deep learning-based inverse problems. But it is quantitatively unclear how good the unrolled models are, among all possible model variations. To fill in this ... | SP:e9cb82d442fd1f42348d33be29e2735da7e13dbe |
Robust Overfitting may be mitigated by properly learned smoothening | 1 INTRODUCTION Adversarial training ( AT ) ( Madry et al. , 2018 ) , i.e. , training a deep network to minimize the worst-case training loss under input perturbations , is recognized as the current best defense method to adversarial attacks . However , one of its pitfalls was exposed by a recent work ( Rice et al. , 20... | The paper studies a method for mitigating robust overfitting. Rice et al., and others have observed that when training a neural network robustly on say CIFAR10, then the robust test error often overfits, i.e., it has a U-shaped curve as a function of training epochs. Rice et al. demonstrated that early stopping the rob... | SP:287426061a33fd5cef9b00660c06e98f3af010d2 |
Trajectory Prediction using Equivariant Continuous Convolution | Trajectory prediction is a critical part of many AI applications , for example , the safe operation of autonomous vehicles . However , current methods are prone to making inconsistent and physically unrealistic predictions . We leverage insights from fluid dynamics to overcome this limitation by considering internal sy... | This paper presents a novel Equivariant Continous COnvolution (ECCO) method for vehicle and pedestrian trajectory prediction. ECCO extends the previous continuous convolution method and makes it rotationally equivariant. To achieve that, they constrain the convolution kernel function K and make only the K(0, r) compone... | SP:7699e91f9cf5149401b1adf811bcdaa643869262 |
Disentangled Generative Causal Representation Learning | 1 INTRODUCTION Consider the observed data x from a distribution qx on X ⊆ Rd and the latent variable z from a prior pz on Z ⊆ Rk . In bidirectional generative models ( BGMs ) , we are normally interested in learning an encoder E : X → Z to infer latent variables and a generator G : Z → X to generate data , to achieve b... | This paper presents a latent variable model where the variables in the latent space are causally disentangled, i.e. the disentanglement is ensured according to a structural causal model (SCM). The resulting model is made up of two parts. The first one, a generative unsupervised part, is essentially a VAE and is defined... | SP:f2f1c3e0201395340e06f5873299639a8f4d16ee |
Parameterized Pseudo-Differential Operators for Graph Convolutional Neural Networks | 1 INTRODUCTION . Convolutional neural networks have performed incredibly well on tasks such as image classification , segmentation , and object detection ( Khan et al. , 2020 ) . While there have been diverse architectural design innovations leading to improved accuracies across these tasks , all of these tasks share t... | The paper defines simple differential operators at nodes in a graph (gradient, Laplacian) and uses them in the proposed graph convolutional layer. The claim is that simple operators will limit the representation power of the layer leading to better generalization. While this might be true at some level, it goes against... | SP:829267dac365b8cdd66c0188da58602135a0c4b9 |
Image GANs meet Differentiable Rendering for Inverse Graphics and Interpretable 3D Neural Rendering | 1 INTRODUCTION . The ability to infer 3D properties such as geometry , texture , material , and light from photographs is key in many domains such as AR/VR , robotics , architecture , and computer vision . Interest in this problem has been explosive , particularly in the past few years , as evidenced by a large body of... | This paper proposes to couple a GAN, an inverse graphics network, and a differentiable renderer. The authors base their work on StyleGAN, and use the observation that a specific part of the latent code corresponds to camera view-point to rapidly annotate a large amount of synthetic images with approximate camera pose. ... | SP:31ecf3efa2c7a0276d2f8fc761d358cfeed6e98d |
Self-supervised Contrastive Zero to Few-shot Learning from Small, Long-tailed Text data | 1 INTRODUCTION . The current prevailing approach to supervised and few-shot learning is to use self-supervised pretraining on large-scale ‘ task-external ’ data and then fine-tune on end-task labels . Recent studies have found that , thus far , this way of pretraining fails in low-resource settings ( Yogatama et al. , ... | This paper proposes a contrastive autoencoder approach that only requires small data to perform a multi-label classification on the long-tail problem. They introduce a matching network to compare text and label embeddings and calculate the probabilities of the label given the input. The proposed idea is very straightfo... | SP:66e2413a5a20b51378742286d20985a11776a782 |
Exploring Routing Strategies for Multilingual Mixture-of-Experts Models | 1 INTRODUCTION . Scaling up neural network models has recently received great attention , given the significant quality improvements in a variety of areas such as natural language understanding ( Raffel et al. , 2019 ; Brown et al. , 2020 ) and multilingual machine translation ( Huang et al. , 2019 ; Lepikhin et al. , ... | In this paper, the authors explore alternatives to the standard token-based routing in sparsely-gated MoE models for multilingual NMT. This exploration is motivated by the need for efficient inference in MoE models, for which token-based routing is a limitation. The alternative is task-based routing, where examples for... | SP:37c923f6c8e655da32a295e69856cf7d7eff9618 |
Bidirectionally Self-Normalizing Neural Networks | 1 INTRODUCTION . Neural networks have brought unprecedented performance in various artificial intelligence tasks ( Graves et al. , 2013 ; Krizhevsky et al. , 2012 ; Silver et al. , 2017 ) . However , despite decades of research , training neural networks is still mostly guided by empirical observations and successful t... | In this paper, the authors introduce the bidirectional self-normalizing neural networks (BSNN) that preserve the norms in both forward and backward passes. To serve such purpose, a new class of activation functions, GPN, is proposed, which can be obtained via the affine transform of existing activation functions like t... | SP:54c5295194b84ff9c33532c6c556558575e42419 |
Segmenting Natural Language Sentences via Lexical Unit Analysis | 1 INTRODUCTION . Sequence segmentation is essentially the process of partitioning a sequence of fine-grained lexical units into a sequence of coarse-grained ones . In some scenarios , each composed unit is assigned a categorical label . For example , Chinese word segmentation splits a character sequence into a word seq... | The paper is well-written, easy to follow and clear. However, the novelty and main contribution of the paper is not clear. The authors used a scoring model to score the composition of each segment, as well as the probability of having a specific label for the segment. The BERT language model is used in the paper to enc... | SP:a8e0b9e55e9a0648ba1c64cf0edb8f09c9a38109 |
Learning Algebraic Representation for Abstract Spatial-Temporal Reasoning | 1 INTRODUCTION . “ Thought is in fact a kind of Algebra. ” —William James ( James , 1891 ) Imagine you are given two alphabetical sequences of “ c , b , a ” and “ d , c , b ” , and asked to fill in the missing element in “ e , d , ? ” . In nearly no time will one realize the answer to be c. However , more surprising fo... | This work proposes a new learner bridging the gap between connectionists and classicists in the task of Raven’s Progressive Matrices (RPM). It relies on a CNN to extract visual features and then uses an algebraic abstract reasoning module to infer the operators of an RPM instance, which allows applying the inferred ope... | SP:258fe1091f7ce89bff79cd8377ee5faad84e9315 |
CTRLsum: Towards Generic Controllable Text Summarization | 1 INTRODUCTION . Neural summarization systems aim to compress a document into a short paragraph or sentence while preserving key information . There are largely two categories of summarization systems : extractive summarization that extracts important portions of a document ( Cheng & Lapata , 2016 ; Nallapati et al. , ... | This paper proposes a two-stage summarization system where a document is provided along with (optionally) keywords or a prompt. This supplemental information helps to guide the summarization and possibly make it more user-specific. The keywords and prompt can also be guessed automatically by a BERT-base model, which se... | SP:a50de9e3cf34fd189763ee172fcff026cbc679dc |
Provable Robustness by Geometric Regularization of ReLU Networks | 1 INTRODUCTION . Neural networks have been very successful in tasks such as image classification and speech recognition . However , recent work ( Szegedy et al. , 2014 ; Goodfellow et al. , 2015 ) has demonstrated that neural networks classifiers can be arbitrarily fooled by small , adversarially-chosen perturbations o... | The authors exploit the piecewise linear nature of ReLU neural networks to design a new regularizer that improves the robustness of the neural network. It can be viewed as a alternative to the regularizer proposed in Croce et al. (2018) -- the current regularizer uses the analytic center, whereas the previous work (nam... | SP:77ff356f24bca397a8f89706e0f89ff14b6b81be |
One Vertex Attack on Graph Neural Networks-based Spatiotemporal Forecasting | 1 INTRODUCTION . Spatiotemporal traffic forecasting has been a long-standing research topic and a fundamental application in intelligent transportation systems ( ITS ) . For instance , with better prediction of future traffic states , navigation apps can help drivers avoid traffic congestion , and traffic signals can m... | This paper studies the problem of attacking graph neural networks for spatio-temporal prediction problems (e.g., traffic speed prediction). The input of the problem is a spatio-temporal sequence represented as graphs at time t-N+1 to t, where a graph neural network is trained to predict the graph sequence for time t+1 ... | SP:a26ff5fe208e5a7a24775d3823d886fc68b89997 |
Understanding Classifiers with Generative Models | 1 INTRODUCTION . Machine learning algorithms have shown remarkable success in challenging supervised learning tasks such as object classification ( He et al. , 2016 ) and speech recognition ( Graves et al. , 2013 ) . Deep neural networks in particular , have gained traction because of their ability to learn a hierarchi... | The authors provide a new method for detecting when deep networks are likely to fail and demonstrate through extensive experimentation its accuracy against generalization errors, out of distribution samples and adversarial attacks. The method builds on prior Mahalanobis metric of (Kimin Lee, et al., A unified framework... | SP:6ae82744b305ffa175e482c92cc79137456bc2ee |
Targeted VAE: Structured Inference and Targeted Learning for Causal Parameter Estimation | Undertaking causal inference with observational data is extremely useful across a wide range of domains including the development of medical treatments , advertisements and marketing , and policy making . There are two main challenges associated with undertaking causal inference using observational data : treatment ass... | The proposed contribution of this work is to build of the existing literature which uses variational autoencoders for causal inference by (1) allowing an explicit mechanism for modeling irrelevant covariates and (2) incorporating targeted regularization into the latent variable nnet framework. Optimization of the model... | SP:85919ada5493c7f63cbd171e7f9738fee02d8dfb |
Why Convolutional Networks Learn Oriented Bandpass Filters: Theory and Empirical Support | It has been repeatedly observed that convolutional architectures when applied to image understanding tasks learn oriented bandpass filters . A standard explanation of this result is that these filters reflect the structure of the images that they have been exposed to during training : Natural images typically are local... | This paper presents an explanation of why convolutional neural networks learn oriented bandpass filters - as has been commonly shown for early layers in various ConvNet architectures. The main argument is that oriented bandpass filters are the eigen-functions of localized convolution operators and in order to span the ... | SP:00a28b287979b3cf803c21118f4e403a92e4f479 |
Discovering a set of policies for the worst case reward | 1 INTRODUCTION . Reinforcement learning ( RL ) is concerned with building agents that can learn to act so as to maximize reward through trial-and-error interaction with the environment . There are several reasons why it can be useful for an agent to learn about multiple ways of behaving , i.e. , learn about multiple po... | Given a rewardless environment MDP, the authors want to find a set of policies for the worst case reward function. Their process involves two steps: first to select the right set of policies and second to combine them to generate a new policy. The policy selection is made with the only goal to maximize the expected ret... | SP:41065df46326876b201c82ec287033ff43e9bcc8 |
Implicit Regularization of SGD via Thermophoresis | A central ingredient in the impressive predictive performance of deep neural networks is optimization via stochastic gradient descent ( SGD ) . While some theoretical progress has been made , the effect of SGD in neural networks is still unclear , especially during the early phase of training . Here we generalize the t... | This paper proposes that SGD has the implicitly bias of reducing gradient variance via the phenomenon of thermophoresis that masses tend to flow from regions with higher temperature / variance of random walk to regions with lower temperature / variance of random walk. In the setup of two-layer neural networks trained b... | SP:b7704e25b5f177afa8f8d85636d652b5079afc0e |
End-to-End on-device Federated Learning: A case study | 1 INTRODUCTION . With the development of computation capability in devices , Machine Learning and Deep Learning arouse great interests by companies who are eager to utilize ML/DL methods to improve their service quality . However , with the explosive growth of data generated on edge devices , the traditional centralize... | This paper applies federated learning to steering wheel prediction for autonomous driving. "Federated learning" in this draft mainly refers to an on-device distributed training algorithm where each edge device hosts its private data and performs local updates (model training) and send the updates back to a central ser... | SP:f0bf4f7a726d20e1ebf8d45033a61e3034ede044 |
Self-Supervised Multi-View Learning via Auto-Encoding 3D Transformations | 1 INTRODUCTION . 3D object representation has become increasingly prominent for a wide range of applications , such as 3D object recognition and retrieval ( Maturana & Scherer , 2015 ; Qi et al. , 2016 ; Brock et al. , 2016 ; Qi et al. , 2017a ; b ; Klokov & Lempitsky , 2017 ; Su et al. , 2015 ; Feng et al. , 2018 ; Yu... | This paper proposed a self-supervised learning method of 3D shape descriptors for 3D recognition through multi-view 2D image representation learning. To represent the 3D shape, the authors first project the object to a group of 2D project images, which helps apply deep learning due to the image's matrix data format. T... | SP:475921dfd2c656b69172acf8d3ac49ecde54639d |
Does injecting linguistic structure into language models lead to better alignment with brain recordings? | 1 INTRODUCTION . Recent advances in deep neural networks for natural language processing ( NLP ) have generated excitement among computational neuroscientists , who aim to model how the brain processes language . These models are argued to better capture the complexity of natural language semantics than previous comput... | of paper: the authors explore adding a soft structural attention constraint to BERT, by penalizing attention weights that are substantially different from a head–dependent "adjacency" matrix derived from dependency parses. BERT is then fine-tuned with and without ("domain-finetuned") this constraint on corpus data for... | SP:38e84c7dbf88091348af8b84192d8383c4b37b5b |
Hopfield Networks is All You Need | We introduce a modern Hopfield network with continuous states and a corresponding update rule . The new Hopfield network can store exponentially ( with the dimension of the associative space ) many patterns , retrieves the pattern with one update , and has exponentially small retrieval errors . It has three types of en... | This work extends the binary Hopfield network (Demircigil et al., 2017) to continuous patterns and states. Connections are drawn between the result model to the attention layers of the transformers, the pooling operation of LSTM, similarity search, and fully connected layers. Experimental results are briefly described ... | SP:1387491705ff05c21b119fad95ef3e63beaa57c9 |
Revisiting Dynamic Convolution via Matrix Decomposition | 1 INTRODUCTION . Dynamic convolution ( Yang et al. , 2019 ; Chen et al. , 2020c ) has recently become popular for the implementation of light-weight networks ( Howard et al. , 2017 ; Zhang et al. , 2018b ) . Its ability to achieve significant performance gains with negligible computational cost has motivated its adopti... | This paper proposes a technique to reduce parameter count and improve training stability for dynamic convolutions using matrix decomposition. It looks at prior work (CondConv, DyConv) which aggregate multiple convolutional kernels via an attention score, and suggests this vanilla formulation is redundant since it sums ... | SP:66b06c7dee715568ea863831c610f1e6ebca05be |
Local Clustering Graph Neural Networks | 1 INTRODUCTION . Recent emergence of the Graph Neural Networks ( GNNs ) , exemplified by models like ChebyNet ( Defferrard et al. , 2016 ) , GCN ( Kipf & Welling , 2017 ) , GraphSAGE ( Hamilton et al. , 2017 ) , GAT ( Veličković et al. , 2018 ) , and GIN ( Xu et al. , 2019 ) , has drastically reshaped the landscape o... | In this paper, the authors study the connection between GNNs and local clustering, and find that short random-walks in GNNs have a high probability to be stuck at a local cluster. Based on this, they propose a light and scalable GNN learning framework called LCGNN, which first adopts the local clustering method PPR-Nib... | SP:8f15ab1eb05ed48f21dd35a118eb299040960074 |
AT-GAN: An Adversarial Generative Model for Non-constrained Adversarial Examples | 1 INTRODUCTION . In recent years , Deep Neural Networks ( DNNs ) have been found vulnerable to adversarial examples ( Szegedy et al. , 2014 ) , which are well-crafted samples with tiny perturbations imperceptible to humans but can fool the learning models . Despite the great success of the deep learning empowered appli... | The paper proposes AT-GAN (Adversarial Transfer on Generative Adversarial Net) to train an adversarial generative model that can directly produce adversarial examples. Different from previous works, the study aims to learn the distribution of adversarial examples so as to generate semantically meaningful adversaries. A... | SP:e9d2702f8ac6f04fd5429fc1236aa163d179a0aa |
FMix: Enhancing Mixed Sample Data Augmentation | 1 INTRODUCTION . Recently , a plethora of approaches to Mixed Sample Data Augmentation ( MSDA ) have been proposed which obtain state-of-the-art results , particularly in classification tasks ( Chawla et al. , 2002 ; Zhang et al. , 2017 ; Tokozume et al. , 2017 ; 2018 ; Inoue , 2018 ; Yun et al. , 2019 ; Takahashi et a... | This paper proposes an advanced masking strategy for CutMix augmentation based on the low-pass filter. The authors provide an interesting mutual information analysis for different augmentation strategies to describe their motivation. The experiments include many vision tasks (CIFAR-10, CIFAR-100, Fashion-MNIST, Tiny-Im... | SP:02ad24f0c92d8f6203be90ff0c173036f76c9959 |
Neural gradients are near-lognormal: improved quantized and sparse training | 1 INTRODUCTION . Neural gradients are used in the training process of deep networks to backpropagate the error-gradient throughout the model , thus allowing to compute the required weight updates . As these neural gradients are needed for a substantial ratio of the underlying computations ( about 23 ) , compressing the... | This work proposed a very interesting idea that the back-propagated errors have log-normal distributions. The authors could extend this intriguing observation into computation efficient algorithms; reduced-precision floating-point quantization or the pruning of the back-prop error, which are very interesting. The autho... | SP:56f796be21ad1e4a563138ba70053071cc419e8c |
Privacy Preserving Recalibration under Domain Shift | 1 INTRODUCTION . Machine learning classifiers are currently deployed in high stakes applications where ( 1 ) the cost of failure is high , so prediction uncertainty must be accurately calibrated ( 2 ) the test distribution does not match the training distribution , and ( 3 ) data is subject to privacy constraints . All... | The paper studies the problem of classifier recalibration under differential privacy constraints. They propose a framework with a calibrator and several private data sources, and it works as follows. At each iteration, the calibrator queries each source, and the data source sends back the private answer, which will be ... | SP:30332615c0031634abb0108b91caad1657a5e8be |
Transformer-QL: A Step Towards Making Transformer Network Quadratically Large | 1 INTRODUCTION . Since its introduction in Vaswani et al . ( 2017 ) , Transformer networks have overtaken its predecessor Recurrent Neural Networks ( RNN ) in almost every natural language processing task . However , one limitation of Transformer network is its high requirement of memory and computational power . In a ... | The work introduces the Transformer-QL, a transformer-based model that aims to capture long distance dependencies in the input. The network processes the information defining multiple temporal scales, with finer scales for nearby elements, and coarser scales for distant information. It also includes the recurrent memor... | SP:5e5fb9699cfc3ee5368b83d473e2e6289e372714 |
Hierarchical Meta Reinforcement Learning for Multi-Task Environments | 1 INTRODUCTION . With great breakthrough of deep reinforcement learning ( DRL ) methods ( Mnih et al. , 2015 ; Silver et al. , 2016 ; Mnih et al. , 2016 ; Schulman et al. , 2015 ; Lillicrap et al. , 2015 ) , it is an urgent need to use DRL methods to solve more complex decision-making problems . The practical problem i... | This paper considers a FPS game that can be decomposed into two sub-tasks, navigation and shooting. A hierarchical meta RL method is introduced and the updating rules for sub-policies and meta parameters are provided. Experiments focus on this specific environment and hence the hierarchical structure is also specified ... | SP:1d1f110bbf38b9ed8faeefa13e59367e2945206c |
Generalizing and Tensorizing Subgraph Search in the Supernet | 1 INTRODUCTION . Deep learning ( Goodfellow et al. , 2017 ) has been successfully applied in many applications , such as image classification for computer vision ( CV ) ( LeCun et al. , 1998 ; Krizhevsky et al. , 2012 ; He et al. , 2016 ; Huang et al. , 2017 ) and language modeling for natural language processing ( NLP... | + This paper generalizes the supernet search problem on a broader horizon. Specifically, some of the current NAS methods use supernet to co-training different neural architectures for further architecture search. This paper does not just consider supernet as a tool for NAS, but also consider supernet as a graphical mod... | SP:1f7be784b1ff0491f0d78b5eefe1a7706036feeb |
$i$-Mix: A Domain-Agnostic Strategy for Contrastive Representation Learning | 1 INTRODUCTION . Representation learning ( Bengio et al. , 2013 ) is a fundamental task in machine learning since the success of machine learning relies on the quality of representation . Self-supervised representation learning ( SSL ) has been successfully applied in several domains , including image recognition ( He ... | The key idea of this paper is to apply MixUp-style regularization to self-supervised contrastive learning techniques (SimCLR, MoCo-v2, BYOL). This is combined with another form of MixUp that involves only the images (not the labels), but the precise nature of this component is unclear. For large networks trained on sma... | SP:f346cf947c90327e475698f0d0018064c2497b64 |
Stochastic Canonical Correlation Analysis: A Riemannian Approach | t ) convergence rate ( where d is the dimensionality ) or only extract the top 1 component with O ( 1t ) convergence rate . In contrast , our algorithm achieves O ( d2k ) runtime complexity per iteration for extracting top k canonical components with O ( 1t ) convergence rate . We present our theoretical analysis as we... | The paper presents an approach to find canonical directions in a streaming fashion, i.e. without direct calculation of covariance matrices (which becomes hard when the number of examples is large). This solution to that task is not obvious, because the objective function of CCA, together with whitening constraints, doe... | SP:1355359d2a6ca8940e4c3fa3f858779f49156d49 |
Information Condensing Active Learning | 1 Introduction . Machine learning models are widely used for a vast array of real world problems . They have been applied successfully in a variety of areas including biology ( Ching et al. , 2018 ) , chemistry ( SanchezLengeling and Aspuru-Guzik , 2018 ) , physics ( Guest et al. , 2018 ) , and materials engineering ( ... | This paper proposes a way to do batch mode model agnostic active learning. In this task, the agent has to query a batch of data points from a set of unlabeled examples for which it will get labels. The paper puts an additional requirement that the algorithm is model-agnostic. The key idea here is to sample a batch of p... | SP:7ee8acfa502077ecac20e98eb665697bf351407c |
OFFER PERSONALIZATION USING TEMPORAL CONVOLUTION NETWORK AND OPTIMIZATION | Lately , personalized marketing has become important for retail/e-retail firms due to significant rise in online shopping and market competition . Increase in online shopping and high market competition has led to an increase in promotional expenditure for online retailers , and hence , rolling out optimal offers has b... | This paper is about optimizing discount offers to individual customers to maximize business value. A temporal convolutional network (TCN) is used to model the customer's purchase probability. This network is then used to estimate the customer's offer-elasticity. Finally a linear program is used to optimize offers acros... | SP:1b0eb87a5d014c94fb41b0e2322bb31ef2a11b78 |
Behavioral Cloning from Noisy Demonstrations | 1 INTRODUCTION . Imitation learning ( IL ) has become a widely used approach to obtain autonomous robotics control systems . IL is often more applicable in real-world problems than reinforcement learning ( RL ) since expert demonstrations are often easier than designing appropriate rewards that RL requires . There have... | This paper studies the problem of imitation an expert from noisy demonstrations without interactions with the environment. It proposes an algorithm, which utilizes an ensemble of behavioral cloning policies and is analogous to the mean shift algorithm, to find the mode from noisy demonstrations. Experimentally, the pro... | SP:072e0bcba8ff2c75ff0ff55576ae77943dada729 |
Cross-Attentional Audio-Visual Fusion for Weakly-Supervised Action Localization | 1 INTRODUCTION . The goal of this paper is to temporally localize actions and events of interest in videos with weaksupervision . In the weakly-supervised setting , only video-level labels are available during the training phase to avoid expensive and time-consuming frame-level annotation . This task is of great import... | The authors propose to mix the independent embeddings of audio and visual data by a set of cross-attention layers to the task of audio-visual event localization. They weigh the features based on the correlation between the representations of both modalities to obtain the output representation. After several layers like... | SP:32798b2b132f08f55733c163090f097aa0b384bf |
EqCo: Equivalent Rules for Self-supervised Contrastive Learning | In this paper , we propose a method , named EqCo ( Equivalent Rules for Contrastive Learning ) , to make self-supervised learning irrelevant to the number of negative samples in the contrastive learning framework . Inspired by the InfoMax principle , we point that the margin term in contrastive loss needs to be adaptiv... | This paper focuses on self-supervised contrastive learning. Previous contrastive learning methods heavily rely on a large number of negative samples. This paper proposed a novel method with an additional margin term, and mathematically investigate the relationship among the margin term, the temperature, and the number ... | SP:c477e488dfaa82ea6698a52c6677b74135fecd12 |
Demystifying Learning of Unsupervised Neural Machine Translation | 1 INTRODUCTION . Unsupervised Neural Machine Translation or UNMT have grown from its infancy ( Artetxe et al. , 2018 ; Lample et al. , 2018a ) to close-to-supervised performance recently on some translation scenarios ( Lample & Conneau , 2019 ; Song et al. , 2019 ) . Early UNMT works ( Artetxe et al. , 2017 ; Lample et... | This paper performs an ablative study on the two components involved in training unsupervised MT systems: 1) back-translation loss, 2) denoising autoencoding loss. It links the reconstruction loss to ELBO (where the q distribution is a back-translation model). It shows that the original loss with both the components is... | SP:4b87425197bb556af13c6aee324bc5ea2b82fc45 |
Large Scale Image Completion via Co-Modulated Generative Adversarial Networks | 1 INTRODUCTION . Generative adversarial networks ( GANs ) have received a great amount of attention in the past few years , during which a fundamental problem emerges from the divergence of development between image-conditional and unconditional GANs . Image-conditional GANs have a wide variety of computer vision appli... | In this paper, the authors propose a general approach for image completion with large-scale missing regions. The key is to combine image-conditional and modulated unconditional generative architectures via co-modulation. The presented approach has demonstrated strong performance in the image painting with large-scale m... | SP:df623838a4d93f4a6c518f23424d5d2ce2cbf704 |
Learning Robust State Abstractions for Hidden-Parameter Block MDPs | 1 INTRODUCTION A key open challenge in AI research that remains is how to train agents that can learn behaviors that generalize across tasks and environments . When there is common structure underlying the tasks , we have seen that multi-task reinforcement learning ( MTRL ) , where the agent learns a set of tasks simul... | This paper studies a family of Markov Decision Process (MDP) models with a low-dimensional unobserved state, called the block MDP. The authors assume that system dynamics of the underlying MDP is sufficiently summarized by a parameter $\theta$. This learning setting could be seen as a combination of Block MDP and the H... | SP:9e72893f6675196c62be20b31e686364f690479a |
Bayesian Learning to Optimize: Quantifying the Optimizer Uncertainty | 1 INTRODUCTION . Computational models of many real-world applications involve optimizing non-convex objective functions . As the non-convex optimization problem is NP-hard , no optimization algorithm ( or optimizer ) could guarantee the global optima in general , and instead , their solutions ’ usefulness ( sometimes b... | The paper considers the question of quantifying the uncertainty that arises from the optimiser used to perform inference in a given model. Taking a Bayesian approach, the aim is to deduce the posterior over the space of optimisers. The form for the posterior is chosen to be a Boltzmann distribution which is then approx... | SP:dc75166137ad902cb0b08966bc25914e0f141c63 |
Global optimality of softmax policy gradient with single hidden layer neural networks in the mean-field regime | 1 INTRODUCTION . In recent years , deep reinforcement learning has revolutionized the world of Artificial Intelligence by outperforming humans in a multitude of highly complex tasks and achieving breakthroughs that were deemed unthinkable at least for the next decade . Spectacular examples of such revolutionary potenti... | This paper studies the asymptotic convergence properties of (population-level) policy gradient methods with two-layer neural networks, softmax parametrization, and entropic regularization, in the mean-field regime. By modelling the hidden layer as a probability distribution over the parameter space, the training dynami... | SP:18121c6a208ea58c09b24e3af951a17b9ed3cbc3 |
Learning to Solve Multi-Robot Task Allocation with a Covariant-Attention based Neural Architecture | 1 INTRODUCTION . In multi-robot task allocation ( MRTA ) problems , we study how to coordinate tasks among a team of cooperative robotic systems such that the decisions are free of conflict and optimize a quantity of interest ( Gerkey & Matarić , 2004 ) . The potential real-world applications of MRTA are immense , con... | This paper presents an interesting idea of using neural-network-based RL to solve a type of vehicle routing problems, where the vehicles are tasked with visiting spatial locations to deliver items, and are subject to load capacity and delivery time constraints. In order to solve this problem, the authors propose an enc... | SP:fbcb2bbd4ca1133e8ae2178e02d3a7393ec4e05d |
A Simple and General Graph Neural Network with Stochastic Message Passing | Graph neural networks ( GNNs ) are emerging machine learning models on graphs.1 One key property behind the expressiveness of existing GNNs is that the learned2 node representations are permutation-equivariant . Though being a desirable prop-3 erty for certain tasks , however , permutation-equivariance prevents GNNs fr... | This paper proposed a proximity-aware graph neural network while maintaining the permutation equivariance property. The proposed model, dubbed as stochastic message passing (SMP), arguments the existing GNNs with stochastic node representations. The author proved the proposed method can model proximity-aware representa... | SP:45941e6abff2f79dd106783302095a6674da5f4a |
DROPS: Deep Retrieval of Physiological Signals via Attribute-specific Clinical Prototypes | 1 INTRODUCTION . Physiological data are being collected at a burgeoning rate . Such growth is driven by the digitization of previous patient records , the presence of novel health monitoring and recording systems , and the recent recommendation to facilitate the exchange of health records ( European Commission , 2019 )... | This paper proposes to tackle the problem of retrieving and clustering physiological signals by learning clinical prototypes via supervised contrastive learning. Three readily available patient attributes, disease, age, and sex, are used to assist the learning. A hard assignment of samples to prototype is proposed, and... | SP:f64ed00548580d2d06e5d8e529894144940ed630 |
Rao-Blackwellizing the Straight-Through Gumbel-Softmax Gradient Estimator | 1 INTRODUCTION . Models with discrete latent variables are common in machine learning . Discrete random variables provide an effective way to parameterize multi-modal distributions , and some domains naturally have latent discrete structure ( e.g , parse trees in NLP ) . Thus , discrete latent variable models can be fo... | The paper presents a new way algorithm to compute the straight-through variant of the Gumbel Softmax gradient estimator. The method does not change the estimator's bias, but provably reduces its variance (with a small overhead, using Rao-blackwellization). The new estimator shows good performance on different tasks, an... | SP:8351823091c5320244dec89c71d3319a103cd6a4 |
Consensus Clustering with Unsupervised Representation Learning | 1 INTRODUCTION . Supervised learning algorithms have shown great progress recently , but generally require a lot of labeled data . However , in many domains ( e.g. , advertising , social platforms , etc . ) , most of the available data are not labeled and manually labeling it is a very labor , time , and cost intensive... | The authors propose a learning-based approach for image clustering. In particular, similarly to recent algorithms fro unsupervised representation learning, such a DeepCluster, they propose to iterate between clustering the images in the feature space of the network and updating the network weights to respect the cluste... | SP:ba35b6f0d26b7a9d982a99116326047f496e5a03 |
DJMix: Unsupervised Task-agnostic Augmentation for Improving Robustness | 1 INTRODUCTION CNNs are the de facto standard components of image recognition tasks and achieve excellent performance . However , CNNs are vulnerable to unseen noise on input images . Such harmful noise includes not only adversarially generated noise ( Szegedy et al. , 2014 ; Goodfellow et al. , 2014 ) , but also natur... | This paper proposes an alternative data augmentation method for robust image classification. Training on augmented data has been shown to improve robustness. The proposed method mixes discretized images with real images and uses a consistency loss to enforce smoothness among those predictions. Experimental results incl... | SP:150de466f594e7fede616959fe98956e300dfefe |
LLBoost: Last Layer Perturbation to Boost Pre-trained Neural Networks | In this work , we present LLBoost , a theoretically-grounded , computationallyefficient method to boost the validation accuracy of pre-trained overparameterized models without impacting the original training accuracy . LLBoost adjusts the last layer of a neural network by adding a term that is orthogonal to the trainin... | This paper proposes LLBoost that enables adjusting the last linear layer without impacting the training accuracy under the assumption that the last linear layer is in an over-parametrized situation. When the last layer is not over-parametrized, LLBoost first applies the low rank approximation to the training feature ma... | SP:470fdc4e61564e09d538f7ecd1225494e08416f2 |
3D Scene Compression through Entropy Penalized Neural Representation Functions | 1 INTRODUCTION . The ability to render 3D scenes from arbitrary viewpoints can be seen as a big step in the evolution of digital multimedia , and has applications such as mixed reality media , graphic effects , design , and simulations . Often such renderings are based on a number of high resolution images taken of som... | This paper proposes to compress nerf models with entropy loss, where instead of directly training nerf model parameters, it trains a new function F which takes some compressed information and decodes to the nerf models. Then it did the same things as nerf, which render scenes in novel views. The authors show that the f... | SP:a044379a14bccab23cf617ef66896ecd78edb6ea |
Latent Space Semi-Supervised Time Series Data Clustering | 1 INTRODUCTION . Time series data can be defined as any data which contains multiple sequentially ordered measurements . Real world examples of time series data are abundant throughout many domains , including finance , weather , and medicine . One common learning task is to partition a set of time series into clusters... | This paper proposes a semi-supervised architecture for clustering time-series data based on Convolutional Autoencoders (AEs). The model combines the regular reconstruction loss usually employed in AEs with two new losses based on the intrinsic clustering evaluation metrics, Silhouette and DBIndex. The experiments show ... | SP:f9b9968f2228687032c18ac27887a3c70cdfbc1d |
A Learning Theoretic Perspective on Local Explainability | 1 INTRODUCTION . There has been a growing interest in interpretable machine learning , which seeks to help people better understand their models . While interpretable machine learning encompasses a wide range of problems , it is a fairly uncontroversial hypothesis that there exists a trade-off between a model ’ s compl... | This paper presents two main theoretical results as its main contributions. First, the authors provide a bound on the model generalization in terms its local explainability. This bound relates the model generalization, the training accuracy, local explainability, and the complexity of the explanations. Second the au... | SP:d15b10f7790bd3370607b620d28b13b790abd1ec |
Benefit of deep learning with non-convex noisy gradient descent: Provable excess risk bound and superiority to kernel methods | √ n ) that is obtained by usual Rademacher complexity analysis . This discrepancy is induced by the non-convex geometry of the model and the noisy gradient descent used for neural network training provably reaches a near global optimal solution even though the loss landscape is highly non-convex . Although the noisy gr... | The paper shows that a two-layer neural network (although an extension to deeper models seem unproblematic) may outperform a class of linear functions in terms of the excess risk learning rate, and in a minimax optimality analysis, and when approximating a target function from the neural network class. The paper essent... | SP:f04f2cdfcdb5478771da0dc4f28df9f694739d3d |
Spherical Motion Dynamics: Learning Dynamics of Neural Network with Normalization, Weight Decay, and SGD | 1 INTRODUCTION AND BACKGROUND . Normalization techniques ( e.g . Batch Normalization ( Ioffe & Szegedy , 2015 ) or its variants ) are one of the most commonly adopted techniques for training deep neural networks ( DNN ) . A typical normalization can be formulated as following : consider a single unit in a neural networ... | The main goal of the paper is to establish theoretically some previous known results that for scale invariant networks the weight norm has a fixed point with ||w||^4=eta/lambda ||\tilde{g}|| . They also discuss the angular update, which because of scale invariance is basically equivalent to arccos (1-eta lambda) |w_t|^... | SP:cc7b030c76352bfec247751d011c0a6d02c8147e |
Watch-And-Help: A Challenge for Social Perception and Human-AI Collaboration | 1 INTRODUCTION . Humans exhibit altruistic behaviors at an early age ( Warneken & Tomasello , 2006 ) . Without much prior experience , children can robustly recognize goals of other people by simply watching them act in an environment , and are able to come up with plans to help them , even in novel scenarios . In cont... | The paper targets to demonstrate social perception and human-AI collaboration in common household activities. It shows the development of a multi-agent virtual environment that is used to test an AI agent’s ability to reason about other agents’ mental states and help them in unfamiliar scenarios. This is performed by p... | SP:9afb51b717b926a92c9f2a1b3dc7aceb960ff80a |
Deployment-Efficient Reinforcement Learning via Model-Based Offline Optimization | 1 INTRODUCTION . Reinforcement learning ( RL ) algorithms have recently demonstrated impressive success in learning behaviors for a variety of sequential decision-making tasks ( Barth-Maron et al. , 2018 ; Hessel et al. , 2018 ; Nachum et al. , 2019 ) . Virtually all of these demonstrations have relied on highly-freque... | This paper proposes a new approach to learning control policies with improved data efficiency and fewer number of data collection sessions (with each session using a different policy). Further, the authors proposed a new concept of “deployment efficiency”, with a new “deployment” referring to using a new policy to inte... | SP:bfe85369cfa71f6b26477f26d133751ac05b0536 |
Representation Learning for Sequence Data with Deep Autoencoding Predictive Components | 1 INTRODUCTION . Self-supervised representation learning methods aim at learning useful and general representations from large amounts of unlabeled data , which can reduce sample complexity for downstream supervised learning . These methods have been widely applied to various domains such as computer vision ( Oord et a... | This paper proposes Deep Autoencoding Predictive Components (DAPC), a self-supervised representation learning approach for sequential data. In this approach, the model learns to maximize the predictive information, which is the mutual information between past and future time windows. In order to avoid degenerate soluti... | SP:1d642e5532adea5cd782f529fed197448e60c458 |
A Hypergradient Approach to Robust Regression without Correspondence | 1 INTRODUCTION . Regression analysis has been widely used in various machine learning applications to infer the the relationship between an explanatory random variable ( i.e. , the input ) X ∈ Rd and a response random variable ( i.e. , the output ) Y ∈ Ro ( Stanton , 2001 ) . In the classical setting , regression is us... | The authors proposed a novel method for regression problems with outliers. The main idea is to first propose a mixed-integer optimization problem for the regression problem and then and the optimization procedure of finding the solutiuon of the problem differentiable, and the objective function of the problem are also ... | SP:d29300f18c72041296b43246711ffdfa1dc6681d |
SoGCN: Second-Order Graph Convolutional Networks | 1 INTRODUCTION . Deep localized convolutional filters have achieved great success in the field of deep learning . In image recognition , the effectiveness of 3 × 3 kernels as the basic building block in Convolutional Neural Networks ( CNNs ) is shown both experimentally and theoretically ( Zhou , 2020 ) . We are inspir... | This is yet another paper on graph convolutional networks (GCNs). The investigated SoGCN is a second-order GCN, thus a special case of high-order GCNs (namely with multi-hop graph kernels), which have has been proposed earlier by many researchers, such as by Defferrard et al. (2016), by Kipf & Welling (2017) and by Abu... | SP:51c6ba2bd4d1dafe78e6da30e18577e16ba4fec9 |
Democratizing Evaluation of Deep Model Interpretability through Consensus | 1 INTRODUCTION . Due to the over-parameterization nature ( Allen-Zhu et al. , 2019 ) , deep neural networks ( DNNs ) ( LeCun et al. , 2015 ) have been widely used to handle machine learning and artificial intelligence tasks , however it is often difficult to understand the prediction results of DNNs despite the very go... | The paper deals with explainable machine learning in the supervised setting and especially tackles the case where no ground truth data for evaluating the generate explanations, such as bounding boxes for objects, is available. The proposed "Concensus" approach retrains established architectures on the target dataset an... | SP:15122fcea632ba9f420bd8a538f708a7621c8323 |
Learning to Generate Noise for Multi-Attack Robustness | 1 INTRODUCTION . Deep neural networks have demonstrated enormous success on multiple benchmark applications ( Amodei et al. , 2016 ; Devlin et al. , 2018 ) , by achieving super-human performance on certain tasks . However , to deploy them to safety-critical applications ( Shen et al. , 2017 ; Chen et al. , 2015 ; Mao e... | The authors propose a new method to improve robustness to adversarial examples under various norms (L1, L2 and LInf). Their method combines adversarial training with an adversarial noise generator. They improve upon adversarial training in a multi norm setting by choosing one norm at random for each sample, instead of ... | SP:203a33205b1cacb84b4d31c5b1b3a5cbb4d93742 |
Modeling the Second Player in Distributionally Robust Optimization | 1 INTRODUCTION . Machine learning models trained with empirical risk minimization ( ERM ) are able to achieve high aggregate performance on data sampled from their training distribution . However , they often exhibit drops in accuracy when confronted with data from domains that are under-represented in their training d... | The paper proposes to define the uncertainty set in the DRO problem as a family of parametric generative models, which is to allow more flexibility in the choice of the uncertainty set architecture. To realize this idea, the paper first proposes a new relaxation of the DRO game's inner maximization problem (with KL con... | SP:a097bea86250950d5b3c5be7676c2b390663098e |
Pseudo Label-Guided Multi Task Learning for Scene Understanding | 1 INTRODUCTION . Scene understanding has become increasingly popular in both academia and industry as an essential technology for realizing a variety of vision-based applications such as robotics and autonomous driving . 3D geometric and semantic information of a scene often serve as a basic building block for high-lev... | The paper presents a joint learning strategy for simultaneous semantic segmentation and monocular depth estimation. The main idea is to exploit stereo pairs in training and introduce pseudo-depth label estimated from pre-trained stereo-matching networks. Given the pseudo-depth with confidence estimation, the method pro... | SP:5ac6f67060ad4fa1470c5f87e8329ef293f2025c |
Catastrophic Fisher Explosion: Early Phase Fisher Matrix Impacts Generalization | 1 INTRODUCTION . Implicit regularization in gradient-based training of deep neural networks ( DNNs ) remains relatively poorly understood , despite being considered a critical component in their empirical success ( Neyshabur , 2017 ; Zhang et al. , 2016 ; Jiang et al. , 2020b ) . Recent work suggests that the early pha... | This paper empirically investigates the effect of the trace of the Fisher Information Matrix (FIM) early in training has on the generalization of SGD. Authors demonstrate that the effect of optimally chosen learning rate and batch size for SGD can be modeled as an implicit penalty on the trace of FIM. They argue that e... | SP:8fb2da71029fc4096f279c5873a2c55e8afaa947 |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.