id
stringlengths
9
10
submitter
stringlengths
1
64
authors
stringlengths
4
20.7k
title
stringlengths
4
246
comments
stringlengths
1
523
journal-ref
stringlengths
4
404
doi
stringlengths
11
153
report-no
stringlengths
2
254
categories
stringlengths
5
98
license
stringclasses
9 values
orig_abstract
stringlengths
14
3.35k
versions
listlengths
1
60
update_date
stringlengths
10
10
authors_parsed
listlengths
1
1.35k
abstract
stringlengths
11
3.34k
1412.6575
Bishan Yang
Bishan Yang, Wen-tau Yih, Xiaodong He, Jianfeng Gao, Li Deng
Embedding Entities and Relations for Learning and Inference in Knowledge Bases
12 pages, 4 figures
null
null
null
cs.CL
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
We consider learning representations of entities and relations in KBs using the neural-embedding approach. We show that most existing models, including NTN (Socher et al., 2013) and TransE (Bordes et al., 2013b), can be generalized under a unified learning framework, where entities are low-dimensional vectors learned from a neural network and relations are bilinear and/or linear mapping functions. Under this framework, we compare a variety of embedding models on the link prediction task. We show that a simple bilinear formulation achieves new state-of-the-art results for the task (achieving a top-10 accuracy of 73.2% vs. 54.7% by TransE on Freebase). Furthermore, we introduce a novel approach that utilizes the learned relation embeddings to mine logical rules such as "BornInCity(a,b) and CityInCountry(b,c) => Nationality(a,c)". We find that embeddings learned from the bilinear objective are particularly good at capturing relational semantics and that the composition of relations is characterized by matrix multiplication. More interestingly, we demonstrate that our embedding-based rule extraction approach successfully outperforms a state-of-the-art confidence-based rule mining approach in mining Horn rules that involve compositional reasoning.
[ { "created": "Sat, 20 Dec 2014 01:37:16 GMT", "version": "v1" }, { "created": "Sat, 27 Dec 2014 00:18:17 GMT", "version": "v2" }, { "created": "Fri, 10 Apr 2015 15:24:59 GMT", "version": "v3" }, { "created": "Sat, 29 Aug 2015 15:08:45 GMT", "version": "v4" } ]
2015-09-01
[ [ "Yang", "Bishan", "" ], [ "Yih", "Wen-tau", "" ], [ "He", "Xiaodong", "" ], [ "Gao", "Jianfeng", "" ], [ "Deng", "Li", "" ] ]
We consider learning representations of entities and relations in KBs using the neural-embedding approach. We show that most existing models, including NTN (Socher et al., 2013) and TransE (Bordes et al., 2013b), can be generalized under a unified learning framework, where entities are low-dimensional vectors learned from a neural network and relations are bilinear and/or linear mapping functions. Under this framework, we compare a variety of embedding models on the link prediction task. We show that a simple bilinear formulation achieves new state-of-the-art results for the task (achieving a top-10 accuracy of 73.2% vs. 54.7% by TransE on Freebase). Furthermore, we introduce a novel approach that utilizes the learned relation embeddings to mine logical rules such as "BornInCity(a,b) and CityInCountry(b,c) => Nationality(a,c)". We find that embeddings learned from the bilinear objective are particularly good at capturing relational semantics and that the composition of relations is characterized by matrix multiplication. More interestingly, we demonstrate that our embedding-based rule extraction approach successfully outperforms a state-of-the-art confidence-based rule mining approach in mining Horn rules that involve compositional reasoning.
2309.05295
Karlis Freivalds
Karlis Freivalds, Emils Ozolins, Guntis Barzdins
Discrete Denoising Diffusion Approach to Integer Factorization
International Conference on Artificial Neural Networks ICANN 2023
null
null
null
cs.LG cs.AI
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
Integer factorization is a famous computational problem unknown whether being solvable in the polynomial time. With the rise of deep neural networks, it is interesting whether they can facilitate faster factorization. We present an approach to factorization utilizing deep neural networks and discrete denoising diffusion that works by iteratively correcting errors in a partially-correct solution. To this end, we develop a new seq2seq neural network architecture, employ relaxed categorical distribution and adapt the reverse diffusion process to cope better with inaccuracies in the denoising step. The approach is able to find factors for integers of up to 56 bits long. Our analysis indicates that investment in training leads to an exponential decrease of sampling steps required at inference to achieve a given success rate, thus counteracting an exponential run-time increase depending on the bit-length.
[ { "created": "Mon, 11 Sep 2023 08:26:08 GMT", "version": "v1" } ]
2023-09-12
[ [ "Freivalds", "Karlis", "" ], [ "Ozolins", "Emils", "" ], [ "Barzdins", "Guntis", "" ] ]
Integer factorization is a famous computational problem unknown whether being solvable in the polynomial time. With the rise of deep neural networks, it is interesting whether they can facilitate faster factorization. We present an approach to factorization utilizing deep neural networks and discrete denoising diffusion that works by iteratively correcting errors in a partially-correct solution. To this end, we develop a new seq2seq neural network architecture, employ relaxed categorical distribution and adapt the reverse diffusion process to cope better with inaccuracies in the denoising step. The approach is able to find factors for integers of up to 56 bits long. Our analysis indicates that investment in training leads to an exponential decrease of sampling steps required at inference to achieve a given success rate, thus counteracting an exponential run-time increase depending on the bit-length.
1404.4997
Eric Price
Moritz Hardt and Eric Price
Tight bounds for learning a mixture of two gaussians
STOC 2015
null
null
null
cs.LG cs.DS stat.ML
http://creativecommons.org/licenses/by-nc-sa/3.0/
We consider the problem of identifying the parameters of an unknown mixture of two arbitrary $d$-dimensional gaussians from a sequence of independent random samples. Our main results are upper and lower bounds giving a computationally efficient moment-based estimator with an optimal convergence rate, thus resolving a problem introduced by Pearson (1894). Denoting by $\sigma^2$ the variance of the unknown mixture, we prove that $\Theta(\sigma^{12})$ samples are necessary and sufficient to estimate each parameter up to constant additive error when $d=1.$ Our upper bound extends to arbitrary dimension $d>1$ up to a (provably necessary) logarithmic loss in $d$ using a novel---yet simple---dimensionality reduction technique. We further identify several interesting special cases where the sample complexity is notably smaller than our optimal worst-case bound. For instance, if the means of the two components are separated by $\Omega(\sigma)$ the sample complexity reduces to $O(\sigma^2)$ and this is again optimal. Our results also apply to learning each component of the mixture up to small error in total variation distance, where our algorithm gives strong improvements in sample complexity over previous work. We also extend our lower bound to mixtures of $k$ Gaussians, showing that $\Omega(\sigma^{6k-2})$ samples are necessary to estimate each parameter up to constant additive error.
[ { "created": "Sat, 19 Apr 2014 23:59:35 GMT", "version": "v1" }, { "created": "Mon, 8 Dec 2014 22:15:35 GMT", "version": "v2" }, { "created": "Sun, 17 May 2015 04:47:58 GMT", "version": "v3" } ]
2015-05-19
[ [ "Hardt", "Moritz", "" ], [ "Price", "Eric", "" ] ]
We consider the problem of identifying the parameters of an unknown mixture of two arbitrary $d$-dimensional gaussians from a sequence of independent random samples. Our main results are upper and lower bounds giving a computationally efficient moment-based estimator with an optimal convergence rate, thus resolving a problem introduced by Pearson (1894). Denoting by $\sigma^2$ the variance of the unknown mixture, we prove that $\Theta(\sigma^{12})$ samples are necessary and sufficient to estimate each parameter up to constant additive error when $d=1.$ Our upper bound extends to arbitrary dimension $d>1$ up to a (provably necessary) logarithmic loss in $d$ using a novel---yet simple---dimensionality reduction technique. We further identify several interesting special cases where the sample complexity is notably smaller than our optimal worst-case bound. For instance, if the means of the two components are separated by $\Omega(\sigma)$ the sample complexity reduces to $O(\sigma^2)$ and this is again optimal. Our results also apply to learning each component of the mixture up to small error in total variation distance, where our algorithm gives strong improvements in sample complexity over previous work. We also extend our lower bound to mixtures of $k$ Gaussians, showing that $\Omega(\sigma^{6k-2})$ samples are necessary to estimate each parameter up to constant additive error.
2009.08936
Majdi Radaideh
Majdi I. Radaideh, Koroush Shirvan
Improving Intelligence of Evolutionary Algorithms Using Experience Share and Replay
10 pages, 4 figures, 2 tables
null
null
null
cs.NE cs.LG
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
We propose PESA, a novel approach combining Particle Swarm Optimisation (PSO), Evolution Strategy (ES), and Simulated Annealing (SA) in a hybrid Algorithm, inspired from reinforcement learning. PESA hybridizes the three algorithms by storing their solutions in a shared replay memory. Next, PESA applies prioritized replay to redistribute data between the three algorithms in frequent form based on their fitness and priority values, which significantly enhances sample diversity and algorithm exploration. Additionally, greedy replay is used implicitly within SA to improve PESA exploitation close to the end of evolution. The validation against 12 high-dimensional continuous benchmark functions shows superior performance by PESA against standalone ES, PSO, and SA, under similar initial starting points, hyperparameters, and number of generations. PESA shows much better exploration behaviour, faster convergence, and ability to find the global optima compared to its standalone counterparts. Given the promising performance, PESA can offer an efficient optimisation option, especially after it goes through additional multiprocessing improvements to handle complex and expensive fitness functions.
[ { "created": "Mon, 10 Aug 2020 17:27:30 GMT", "version": "v1" } ]
2020-09-21
[ [ "Radaideh", "Majdi I.", "" ], [ "Shirvan", "Koroush", "" ] ]
We propose PESA, a novel approach combining Particle Swarm Optimisation (PSO), Evolution Strategy (ES), and Simulated Annealing (SA) in a hybrid Algorithm, inspired from reinforcement learning. PESA hybridizes the three algorithms by storing their solutions in a shared replay memory. Next, PESA applies prioritized replay to redistribute data between the three algorithms in frequent form based on their fitness and priority values, which significantly enhances sample diversity and algorithm exploration. Additionally, greedy replay is used implicitly within SA to improve PESA exploitation close to the end of evolution. The validation against 12 high-dimensional continuous benchmark functions shows superior performance by PESA against standalone ES, PSO, and SA, under similar initial starting points, hyperparameters, and number of generations. PESA shows much better exploration behaviour, faster convergence, and ability to find the global optima compared to its standalone counterparts. Given the promising performance, PESA can offer an efficient optimisation option, especially after it goes through additional multiprocessing improvements to handle complex and expensive fitness functions.
1210.4081
Bogdan Savchynskyy
Bogdan Savchynskyy and Stefan Schmidt
Getting Feasible Variable Estimates From Infeasible Ones: MRF Local Polytope Study
20 page, 4 figures
null
null
null
cs.NA cs.CV cs.DS cs.LG math.OC
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
This paper proposes a method for construction of approximate feasible primal solutions from dual ones for large-scale optimization problems possessing certain separability properties. Whereas infeasible primal estimates can typically be produced from (sub-)gradients of the dual function, it is often not easy to project them to the primal feasible set, since the projection itself has a complexity comparable to the complexity of the initial problem. We propose an alternative efficient method to obtain feasibility and show that its properties influencing the convergence to the optimum are similar to the properties of the Euclidean projection. We apply our method to the local polytope relaxation of inference problems for Markov Random Fields and demonstrate its superiority over existing methods.
[ { "created": "Mon, 15 Oct 2012 15:55:34 GMT", "version": "v1" } ]
2012-10-16
[ [ "Savchynskyy", "Bogdan", "" ], [ "Schmidt", "Stefan", "" ] ]
This paper proposes a method for construction of approximate feasible primal solutions from dual ones for large-scale optimization problems possessing certain separability properties. Whereas infeasible primal estimates can typically be produced from (sub-)gradients of the dual function, it is often not easy to project them to the primal feasible set, since the projection itself has a complexity comparable to the complexity of the initial problem. We propose an alternative efficient method to obtain feasibility and show that its properties influencing the convergence to the optimum are similar to the properties of the Euclidean projection. We apply our method to the local polytope relaxation of inference problems for Markov Random Fields and demonstrate its superiority over existing methods.
1804.03082
Hadi Kazemi
Hadi Kazemi, Sobhan Soleymani, Ali Dabouei, Mehdi Iranmanesh, Nasser M. Nasrabadi
Attribute-Centered Loss for Soft-Biometrics Guided Face Sketch-Photo Recognition
Accepted as a conference paper on CVPRW 2018
null
null
null
cs.CV
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
Face sketches are able to capture the spatial topology of a face while lacking some facial attributes such as race, skin, or hair color. Existing sketch-photo recognition approaches have mostly ignored the importance of facial attributes. In this paper, we propose a new loss function, called attribute-centered loss, to train a Deep Coupled Convolutional Neural Network (DCCNN) for the facial attribute guided sketch to photo matching. Specifically, an attribute-centered loss is proposed which learns several distinct centers, in a shared embedding space, for photos and sketches with different combinations of attributes. The DCCNN simultaneously is trained to map photos and pairs of testified attributes and corresponding forensic sketches around their associated centers, while preserving the spatial topology information. Importantly, the centers learn to keep a relative distance from each other, related to their number of contradictory attributes. Extensive experiments are performed on composite (E-PRIP) and semi-forensic (IIIT-D Semi-forensic) databases. The proposed method significantly outperforms the state-of-the-art.
[ { "created": "Mon, 9 Apr 2018 16:15:50 GMT", "version": "v1" } ]
2018-04-10
[ [ "Kazemi", "Hadi", "" ], [ "Soleymani", "Sobhan", "" ], [ "Dabouei", "Ali", "" ], [ "Iranmanesh", "Mehdi", "" ], [ "Nasrabadi", "Nasser M.", "" ] ]
Face sketches are able to capture the spatial topology of a face while lacking some facial attributes such as race, skin, or hair color. Existing sketch-photo recognition approaches have mostly ignored the importance of facial attributes. In this paper, we propose a new loss function, called attribute-centered loss, to train a Deep Coupled Convolutional Neural Network (DCCNN) for the facial attribute guided sketch to photo matching. Specifically, an attribute-centered loss is proposed which learns several distinct centers, in a shared embedding space, for photos and sketches with different combinations of attributes. The DCCNN simultaneously is trained to map photos and pairs of testified attributes and corresponding forensic sketches around their associated centers, while preserving the spatial topology information. Importantly, the centers learn to keep a relative distance from each other, related to their number of contradictory attributes. Extensive experiments are performed on composite (E-PRIP) and semi-forensic (IIIT-D Semi-forensic) databases. The proposed method significantly outperforms the state-of-the-art.
1706.10076
Pavel Kucherbaev
Pavel Kucherbaev, Achilleas Psyllidis, Alessandro Bozzon
Chatbots as Conversational Recommender Systems in Urban Contexts
2 pages, 1 figure, 1 table
null
null
null
cs.SI cs.CY cs.IR
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
In this paper, we outline the vision of chatbots that facilitate the interaction between citizens and policy-makers at the city scale. We report the results of a co-design session attended by more than 60 participants. We give an outlook of how some challenges associated with such chatbot systems could be addressed in the future.
[ { "created": "Fri, 30 Jun 2017 09:24:39 GMT", "version": "v1" }, { "created": "Wed, 9 Aug 2017 09:03:49 GMT", "version": "v2" } ]
2017-08-10
[ [ "Kucherbaev", "Pavel", "" ], [ "Psyllidis", "Achilleas", "" ], [ "Bozzon", "Alessandro", "" ] ]
In this paper, we outline the vision of chatbots that facilitate the interaction between citizens and policy-makers at the city scale. We report the results of a co-design session attended by more than 60 participants. We give an outlook of how some challenges associated with such chatbot systems could be addressed in the future.
1907.05609
Qianwen Wang
Qianwen Wang, Zhen Li, Siwei Fu, Weiwei Cui, Huamin Qu
Narvis: Authoring Narrative Slideshows for Introducing Data Visualization Designs
9 pages, published at IEEE InfoVis 2018,
IEEE Transactions on Visualization and Computer Graphics, vol. 25, no. 1, pp. 779-788, Jan. 2019
10.1109/TVCG.2018.2865232
null
cs.HC
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
Visual designs can be complex in modern data visualization systems, which poses special challenges for explaining them to the non-experts. However, few if any presentation tools are tailored for this purpose. In this study, we present Narvis, a slideshow authoring tool designed for introducing data visualizations to non-experts. Narvis targets two types of end-users: teachers, experts in data visualization who produce tutorials for explaining a data visualization, and students, non-experts who try to understand visualization designs through tutorials. We present an analysis of requirements through close discussions with the two types of end-users. The resulting considerations guide the design and implementation of Narvis. Additionally, to help teachers better organize their introduction slideshows, we specify a data visualization as a hierarchical combination of components, which are automatically detected and extracted by Narvis. The teachers craft an introduction slideshow through first organizing these components, and then explaining them sequentially. A series of templates are provided for adding annotations and animations to improve efficiency during the authoring process. We evaluate Narvis through a qualitative analysis of the authoring experience, and a preliminary evaluation of the generated slideshows.
[ { "created": "Fri, 12 Jul 2019 08:14:18 GMT", "version": "v1" } ]
2019-08-21
[ [ "Wang", "Qianwen", "" ], [ "Li", "Zhen", "" ], [ "Fu", "Siwei", "" ], [ "Cui", "Weiwei", "" ], [ "Qu", "Huamin", "" ] ]
Visual designs can be complex in modern data visualization systems, which poses special challenges for explaining them to the non-experts. However, few if any presentation tools are tailored for this purpose. In this study, we present Narvis, a slideshow authoring tool designed for introducing data visualizations to non-experts. Narvis targets two types of end-users: teachers, experts in data visualization who produce tutorials for explaining a data visualization, and students, non-experts who try to understand visualization designs through tutorials. We present an analysis of requirements through close discussions with the two types of end-users. The resulting considerations guide the design and implementation of Narvis. Additionally, to help teachers better organize their introduction slideshows, we specify a data visualization as a hierarchical combination of components, which are automatically detected and extracted by Narvis. The teachers craft an introduction slideshow through first organizing these components, and then explaining them sequentially. A series of templates are provided for adding annotations and animations to improve efficiency during the authoring process. We evaluate Narvis through a qualitative analysis of the authoring experience, and a preliminary evaluation of the generated slideshows.
1705.08218
Xiaojian Wu
Xiaojian Wu, Yexiang Xue, Bart Selman, Carla P. Gomes
XOR-Sampling for Network Design with Correlated Stochastic Events
In Proceedings of the Twenty-sixth International Joint Conference on Artificial Intelligence (IJCAI-17). The first two authors contribute equally
null
null
null
cs.AI
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
Many network optimization problems can be formulated as stochastic network design problems in which edges are present or absent stochastically. Furthermore, protective actions can guarantee that edges will remain present. We consider the problem of finding the optimal protection strategy under a budget limit in order to maximize some connectivity measurements of the network. Previous approaches rely on the assumption that edges are independent. In this paper, we consider a more realistic setting where multiple edges are not independent due to natural disasters or regional events that make the states of multiple edges stochastically correlated. We use Markov Random Fields to model the correlation and define a new stochastic network design framework. We provide a novel algorithm based on Sample Average Approximation (SAA) coupled with a Gibbs or XOR sampler. The experimental results on real road network data show that the policies produced by SAA with the XOR sampler have higher quality and lower variance compared to SAA with Gibbs sampler.
[ { "created": "Tue, 23 May 2017 12:50:36 GMT", "version": "v1" }, { "created": "Wed, 24 May 2017 01:38:57 GMT", "version": "v2" } ]
2017-05-25
[ [ "Wu", "Xiaojian", "" ], [ "Xue", "Yexiang", "" ], [ "Selman", "Bart", "" ], [ "Gomes", "Carla P.", "" ] ]
Many network optimization problems can be formulated as stochastic network design problems in which edges are present or absent stochastically. Furthermore, protective actions can guarantee that edges will remain present. We consider the problem of finding the optimal protection strategy under a budget limit in order to maximize some connectivity measurements of the network. Previous approaches rely on the assumption that edges are independent. In this paper, we consider a more realistic setting where multiple edges are not independent due to natural disasters or regional events that make the states of multiple edges stochastically correlated. We use Markov Random Fields to model the correlation and define a new stochastic network design framework. We provide a novel algorithm based on Sample Average Approximation (SAA) coupled with a Gibbs or XOR sampler. The experimental results on real road network data show that the policies produced by SAA with the XOR sampler have higher quality and lower variance compared to SAA with Gibbs sampler.
2402.17104
Robert Bassett
Robert L. Bassett, Austin Van Dellen, Anthony P. Austin
Adversarial Perturbations of Physical Signals
null
null
null
null
cs.LG cs.CR eess.SP math.OC stat.ML
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
We investigate the vulnerability of computer-vision-based signal classifiers to adversarial perturbations of their inputs, where the signals and perturbations are subject to physical constraints. We consider a scenario in which a source and interferer emit signals that propagate as waves to a detector, which attempts to classify the source by analyzing the spectrogram of the signal it receives using a pre-trained neural network. By solving PDE-constrained optimization problems, we construct interfering signals that cause the detector to misclassify the source even though the perturbations to the spectrogram of the received signal are nearly imperceptible. Though such problems can have millions of decision variables, we introduce methods to solve them efficiently. Our experiments demonstrate that one can compute effective and physically realizable adversarial perturbations for a variety of machine learning models under various physical conditions.
[ { "created": "Tue, 27 Feb 2024 00:41:00 GMT", "version": "v1" } ]
2024-02-28
[ [ "Bassett", "Robert L.", "" ], [ "Van Dellen", "Austin", "" ], [ "Austin", "Anthony P.", "" ] ]
We investigate the vulnerability of computer-vision-based signal classifiers to adversarial perturbations of their inputs, where the signals and perturbations are subject to physical constraints. We consider a scenario in which a source and interferer emit signals that propagate as waves to a detector, which attempts to classify the source by analyzing the spectrogram of the signal it receives using a pre-trained neural network. By solving PDE-constrained optimization problems, we construct interfering signals that cause the detector to misclassify the source even though the perturbations to the spectrogram of the received signal are nearly imperceptible. Though such problems can have millions of decision variables, we introduce methods to solve them efficiently. Our experiments demonstrate that one can compute effective and physically realizable adversarial perturbations for a variety of machine learning models under various physical conditions.
1007.3353
Laurent Hubert
Laurent Hubert (INRIA - IRISA), Nicolas Barr\'e (INRIA - IRISA), Fr\'ed\'eric Besson (INRIA - IRISA), Delphine Demange (INRIA - IRISA), Thomas Jensen (INRIA - IRISA), Vincent Monfort (INRIA - IRISA), David Pichardie (INRIA - IRISA), Tiphaine Turpin (INRIA - IRISA)
Sawja: Static Analysis Workshop for Java
null
The International Conference on Formal Verification of Object-Oriented Software 2010.13 (2010) 253--267
10.1007/978-3-642-18070-5_7
null
cs.PL
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
Static analysis is a powerful technique for automatic verification of programs but raises major engineering challenges when developing a full-fledged analyzer for a realistic language such as Java. This paper describes the Sawja library: a static analysis framework fully compliant with Java 6 which provides OCaml modules for efficiently manipulating Java bytecode programs. We present the main features of the library, including (i) efficient functional data-structures for representing program with implicit sharing and lazy parsing, (ii) an intermediate stack-less representation, and (iii) fast computation and manipulation of complete programs.
[ { "created": "Tue, 20 Jul 2010 07:03:59 GMT", "version": "v1" } ]
2015-05-19
[ [ "Hubert", "Laurent", "", "INRIA - IRISA" ], [ "Barré", "Nicolas", "", "INRIA - IRISA" ], [ "Besson", "Frédéric", "", "INRIA - IRISA" ], [ "Demange", "Delphine", "", "INRIA - IRISA" ], [ "Jensen", "Thomas", "", "INRIA - IRISA" ], [ "Monfort", "Vincent", "", "INRIA - IRISA" ], [ "Pichardie", "David", "", "INRIA - IRISA" ], [ "Turpin", "Tiphaine", "", "INRIA - IRISA" ] ]
Static analysis is a powerful technique for automatic verification of programs but raises major engineering challenges when developing a full-fledged analyzer for a realistic language such as Java. This paper describes the Sawja library: a static analysis framework fully compliant with Java 6 which provides OCaml modules for efficiently manipulating Java bytecode programs. We present the main features of the library, including (i) efficient functional data-structures for representing program with implicit sharing and lazy parsing, (ii) an intermediate stack-less representation, and (iii) fast computation and manipulation of complete programs.
1804.06926
Leyuan Wang
Leyuan Wang, Yangzihao Wang, Carl Yang and John D. Owens
A Comparative Study on Exact Triangle Counting Algorithms on the GPU
7 pages, 6 figures and 2 tables
null
10.1145/2915516.2915521
null
cs.DC
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
We implement exact triangle counting in graphs on the GPU using three different methodologies: subgraph matching to a triangle pattern; programmable graph analytics, with a set-intersection approach; and a matrix formulation based on sparse matrix-matrix multiplies. All three deliver best-of-class performance over CPU implementations and over comparable GPU implementations, with the graph-analytic approach achieving the best performance due to its ability to exploit efficient filtering steps to remove unnecessary work and its high-performance set-intersection core.
[ { "created": "Wed, 18 Apr 2018 21:51:59 GMT", "version": "v1" } ]
2018-04-20
[ [ "Wang", "Leyuan", "" ], [ "Wang", "Yangzihao", "" ], [ "Yang", "Carl", "" ], [ "Owens", "John D.", "" ] ]
We implement exact triangle counting in graphs on the GPU using three different methodologies: subgraph matching to a triangle pattern; programmable graph analytics, with a set-intersection approach; and a matrix formulation based on sparse matrix-matrix multiplies. All three deliver best-of-class performance over CPU implementations and over comparable GPU implementations, with the graph-analytic approach achieving the best performance due to its ability to exploit efficient filtering steps to remove unnecessary work and its high-performance set-intersection core.
1611.03841
Jie Xu
Jie Xu, Lixing Chen, Kun Liu, Cong Shen
Designing Security-Aware Incentives for Computation Offloading via Device-to-Device Communication
null
null
null
null
cs.GT cs.NI
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
Computation offloading via device-to-device (D2D) communication, or D2D offloading, has recently been proposed to enhance mobile computing performance by exploiting spare computing resources of nearby user devices. The success of D2D offloading relies on user participation in collaborative service provisioning, which incurs extra costs to users providing the service, thus mandating an incentive mechanism that can compensate for these costs. Although incentive mechanism design has been intensively studied in the literature, this paper considers a much more challenging yet less investigated problem in which selfish users are also facing interdependent security risks, such as infectious proximity-based attacks. Security cost is significantly different in nature from conventional service provisioning costs such as energy consumption, since security risks often depend on the collective behavior of all users. To this end, we build a novel mathematical framework by leveraging the combined power of game theory and epidemic theory to investigate the interplay between user incentives and interdependent security risks in D2D offloading, thereby enabling the design of security-aware incentive mechanisms. Our analysis discovers an interesting "less is more" phenomenon: although giving users more incentives promotes more participation, it may harm the network operator's utility. This is because too much participation may foster persistent security risks and as a result, the effective participation level does not improve. Our model and analysis shed new insights on the optimization of D2D offloading networks in the presence of interdependent security risks. Extensive simulations are carried out to verify our analytical conclusions.
[ { "created": "Fri, 11 Nov 2016 20:22:25 GMT", "version": "v1" }, { "created": "Fri, 22 Sep 2017 02:40:12 GMT", "version": "v2" } ]
2017-09-25
[ [ "Xu", "Jie", "" ], [ "Chen", "Lixing", "" ], [ "Liu", "Kun", "" ], [ "Shen", "Cong", "" ] ]
Computation offloading via device-to-device (D2D) communication, or D2D offloading, has recently been proposed to enhance mobile computing performance by exploiting spare computing resources of nearby user devices. The success of D2D offloading relies on user participation in collaborative service provisioning, which incurs extra costs to users providing the service, thus mandating an incentive mechanism that can compensate for these costs. Although incentive mechanism design has been intensively studied in the literature, this paper considers a much more challenging yet less investigated problem in which selfish users are also facing interdependent security risks, such as infectious proximity-based attacks. Security cost is significantly different in nature from conventional service provisioning costs such as energy consumption, since security risks often depend on the collective behavior of all users. To this end, we build a novel mathematical framework by leveraging the combined power of game theory and epidemic theory to investigate the interplay between user incentives and interdependent security risks in D2D offloading, thereby enabling the design of security-aware incentive mechanisms. Our analysis discovers an interesting "less is more" phenomenon: although giving users more incentives promotes more participation, it may harm the network operator's utility. This is because too much participation may foster persistent security risks and as a result, the effective participation level does not improve. Our model and analysis shed new insights on the optimization of D2D offloading networks in the presence of interdependent security risks. Extensive simulations are carried out to verify our analytical conclusions.
2403.09070
Yuxuan Zhao
Yuxuan Zhao, Peiyu Liao, Siting Liu, Jiaxi Jiang, Yibo Lin, Bei Yu
Analytical Heterogeneous Die-to-Die 3D Placement with Macros
null
null
null
null
cs.AR
http://creativecommons.org/licenses/by-nc-nd/4.0/
This paper presents an innovative approach to 3D mixed-size placement in heterogeneous face-to-face (F2F) bonded 3D ICs. We propose an analytical framework that utilizes a dedicated density model and a bistratal wirelength model, effectively handling macros and standard cells in a 3D solution space. A novel 3D preconditioner is developed to resolve the topological and physical gap between macros and standard cells. Additionally, we propose a mixed-integer linear programming (MILP) formulation for macro rotation to optimize wirelength. Our framework is implemented with full-scale GPU acceleration, leveraging an adaptive 3D density accumulation algorithm and an incremental wirelength gradient algorithm. Experimental results on ICCAD 2023 contest benchmarks demonstrate that our framework can achieve 5.9% quality score improvement compared to the first-place winner with 4.0x runtime speedup. Additional experiments on modern RISC-V designs further validate the generalizability and superiority of our framework.
[ { "created": "Thu, 14 Mar 2024 03:26:08 GMT", "version": "v1" }, { "created": "Tue, 13 Aug 2024 13:00:42 GMT", "version": "v2" } ]
2024-08-14
[ [ "Zhao", "Yuxuan", "" ], [ "Liao", "Peiyu", "" ], [ "Liu", "Siting", "" ], [ "Jiang", "Jiaxi", "" ], [ "Lin", "Yibo", "" ], [ "Yu", "Bei", "" ] ]
This paper presents an innovative approach to 3D mixed-size placement in heterogeneous face-to-face (F2F) bonded 3D ICs. We propose an analytical framework that utilizes a dedicated density model and a bistratal wirelength model, effectively handling macros and standard cells in a 3D solution space. A novel 3D preconditioner is developed to resolve the topological and physical gap between macros and standard cells. Additionally, we propose a mixed-integer linear programming (MILP) formulation for macro rotation to optimize wirelength. Our framework is implemented with full-scale GPU acceleration, leveraging an adaptive 3D density accumulation algorithm and an incremental wirelength gradient algorithm. Experimental results on ICCAD 2023 contest benchmarks demonstrate that our framework can achieve 5.9% quality score improvement compared to the first-place winner with 4.0x runtime speedup. Additional experiments on modern RISC-V designs further validate the generalizability and superiority of our framework.
2407.02856
Adrian Pekar
Adrian Pekar and Richard Jozsa
Early-Stage Anomaly Detection: A Study of Model Performance on Complete vs. Partial Flows
9 pages, 5 tables, 2 figures
null
null
null
cs.LG cs.CR
http://creativecommons.org/licenses/by/4.0/
This study investigates the efficacy of machine learning models, specifically Random Forest, in anomaly detection systems when trained on complete flow records and tested on partial flow data. We explore the performance disparity that arises when models are applied to incomplete data typical in real-world, real-time network environments. Our findings demonstrate a significant decline in model performance, with precision and recall dropping by up to 30\% under certain conditions when models trained on complete flows are tested against partial flows. Conversely, models trained and tested on consistently complete or partial datasets maintain robustness, highlighting the importance of dataset consistency in training. The study reveals that a minimum of 7 packets in the test set is required for maintaining reliable detection rates. These results underscore the need for tailored training strategies that can effectively adapt to the dynamics of partial data, enhancing the practical applicability of anomaly detection systems in operational settings.
[ { "created": "Wed, 3 Jul 2024 07:14:25 GMT", "version": "v1" } ]
2024-07-04
[ [ "Pekar", "Adrian", "" ], [ "Jozsa", "Richard", "" ] ]
This study investigates the efficacy of machine learning models, specifically Random Forest, in anomaly detection systems when trained on complete flow records and tested on partial flow data. We explore the performance disparity that arises when models are applied to incomplete data typical in real-world, real-time network environments. Our findings demonstrate a significant decline in model performance, with precision and recall dropping by up to 30\% under certain conditions when models trained on complete flows are tested against partial flows. Conversely, models trained and tested on consistently complete or partial datasets maintain robustness, highlighting the importance of dataset consistency in training. The study reveals that a minimum of 7 packets in the test set is required for maintaining reliable detection rates. These results underscore the need for tailored training strategies that can effectively adapt to the dynamics of partial data, enhancing the practical applicability of anomaly detection systems in operational settings.
1106.1803
H. Blockeel
H. Blockeel, L. Dehaspe, B. Demoen, G. Janssens, J. Ramon, H. Vandecasteele
Improving the Efficiency of Inductive Logic Programming Through the Use of Query Packs
null
Journal Of Artificial Intelligence Research, Volume 16, pages 135-166, 2002
10.1613/jair.924
null
cs.AI
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
Inductive logic programming, or relational learning, is a powerful paradigm for machine learning or data mining. However, in order for ILP to become practically useful, the efficiency of ILP systems must improve substantially. To this end, the notion of a query pack is introduced: it structures sets of similar queries. Furthermore, a mechanism is described for executing such query packs. A complexity analysis shows that considerable efficiency improvements can be achieved through the use of this query pack execution mechanism. This claim is supported by empirical results obtained by incorporating support for query pack execution in two existing learning systems.
[ { "created": "Thu, 9 Jun 2011 13:19:53 GMT", "version": "v1" } ]
2011-06-10
[ [ "Blockeel", "H.", "" ], [ "Dehaspe", "L.", "" ], [ "Demoen", "B.", "" ], [ "Janssens", "G.", "" ], [ "Ramon", "J.", "" ], [ "Vandecasteele", "H.", "" ] ]
Inductive logic programming, or relational learning, is a powerful paradigm for machine learning or data mining. However, in order for ILP to become practically useful, the efficiency of ILP systems must improve substantially. To this end, the notion of a query pack is introduced: it structures sets of similar queries. Furthermore, a mechanism is described for executing such query packs. A complexity analysis shows that considerable efficiency improvements can be achieved through the use of this query pack execution mechanism. This claim is supported by empirical results obtained by incorporating support for query pack execution in two existing learning systems.
1609.03938
Erel Segal-Halevi
Erel Segal-Halevi, Shmuel Nitzan, Avinatan Hassidim, Yonatan Aumann
Envy-Free Division of Land
A preliminary version named 'Envy-free cake-cutting in two dimensions' appeared in the proceedings of AAAI 2015 (https://www.aaai.org/ocs/index.php/AAAI/AAAI15/paper/viewPaper/9656). The main additions here are: (a) handling multi-dimensional resources of arbitrary shape rather than just rectangles, (b) handling an arbitrary number n of agents rather than just 2 or 3, (c) rewriting most proofs
null
10.1287/moor.2019.1016
null
cs.GT cs.CG
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
Classic cake-cutting algorithms enable people with different preferences to divide among them a heterogeneous resource (``cake''), such that the resulting division is fair according to each agent's individual preferences. However, these algorithms either ignore the geometry of the resource altogether, or assume it is one-dimensional. In practice, it is often required to divide multi-dimensional resources, such as land-estates or advertisement spaces in print or electronic media. In such cases, the geometric shape of the allotted piece is of crucial importance. For example, when building houses or designing advertisements, in order to be useful, the allotments should be squares or rectangles with bounded aspect-ratio. We thus introduce the problem of fair land division --- fair division of a multi-dimensional resource wherein the allocated piece must have a pre-specified geometric shape. We present constructive division algorithms that satisfy the two most prominent fairness criteria, namely envy-freeness and proportionality. In settings where proportionality cannot be achieved due to the geometric constraints, our algorithms provide a partially-proportional division, guaranteeing that the fraction allocated to each agent be at least a certain positive constant. We prove that in many natural settings the envy-freeness requirement is compatible with the best attainable partial-proportionality.
[ { "created": "Tue, 13 Sep 2016 17:07:32 GMT", "version": "v1" }, { "created": "Sat, 9 Mar 2019 19:22:15 GMT", "version": "v2" } ]
2021-08-06
[ [ "Segal-Halevi", "Erel", "" ], [ "Nitzan", "Shmuel", "" ], [ "Hassidim", "Avinatan", "" ], [ "Aumann", "Yonatan", "" ] ]
Classic cake-cutting algorithms enable people with different preferences to divide among them a heterogeneous resource (``cake''), such that the resulting division is fair according to each agent's individual preferences. However, these algorithms either ignore the geometry of the resource altogether, or assume it is one-dimensional. In practice, it is often required to divide multi-dimensional resources, such as land-estates or advertisement spaces in print or electronic media. In such cases, the geometric shape of the allotted piece is of crucial importance. For example, when building houses or designing advertisements, in order to be useful, the allotments should be squares or rectangles with bounded aspect-ratio. We thus introduce the problem of fair land division --- fair division of a multi-dimensional resource wherein the allocated piece must have a pre-specified geometric shape. We present constructive division algorithms that satisfy the two most prominent fairness criteria, namely envy-freeness and proportionality. In settings where proportionality cannot be achieved due to the geometric constraints, our algorithms provide a partially-proportional division, guaranteeing that the fraction allocated to each agent be at least a certain positive constant. We prove that in many natural settings the envy-freeness requirement is compatible with the best attainable partial-proportionality.
2404.17428
Manuel Dubinsky
Manuel Dubinsky, Kun-Mao Chao, C\'esar Massri, Gabriel Taubin
Lower Bounds for the Minimum Spanning Tree Cycle Intersection Problem
arXiv admin note: substantial text overlap with arXiv:2301.07643
null
null
null
cs.DM
http://creativecommons.org/licenses/by/4.0/
Minimum spanning trees are important tools in the analysis and design of networks. Many practical applications require their computation, ranging from biology and linguistics to economy and telecommunications. The set of cycles of a network has a vector space structure. Given a spanning tree, the set of non-tree edges defines cycles that determine a basis. The intersection of two such cycles is the number of edges they have in common and the intersection number -- denoted $\cap(G)$ -- is the number of non-empty pairwise intersections of the cycles of the basis. The Minimum Spanning Tree Cycle Intersection problem consists in finding a spanning tree such that the intersection number is minimum. This problem is relevant in order to integrate discrete differential forms. In this paper, we present two lower bounds of the intersection number of an arbitrary connected graph $G=(V,E)$. In the first part, we prove the following statement: $$\frac{1}{2}\left(\frac{\nu^2}{n-1} - \nu\right) \leq \cap(G),$$ where $n = |V|$ and $\nu$ is the \emph{cyclomatic number} of $G$. In the second part, based on some experimental results and a new observation, we conjecture the following improved tight lower bound: $$(n-1) \binom{q}{2} + q \ r\leq \cap(G),$$ where $2 \nu = q (n-1) + r$ is the integer division of $2 \nu$ and $n-1$. This is the first result in a general context, that is for an arbitrary connected graph.
[ { "created": "Fri, 26 Apr 2024 14:08:36 GMT", "version": "v1" } ]
2024-04-29
[ [ "Dubinsky", "Manuel", "" ], [ "Chao", "Kun-Mao", "" ], [ "Massri", "César", "" ], [ "Taubin", "Gabriel", "" ] ]
Minimum spanning trees are important tools in the analysis and design of networks. Many practical applications require their computation, ranging from biology and linguistics to economy and telecommunications. The set of cycles of a network has a vector space structure. Given a spanning tree, the set of non-tree edges defines cycles that determine a basis. The intersection of two such cycles is the number of edges they have in common and the intersection number -- denoted $\cap(G)$ -- is the number of non-empty pairwise intersections of the cycles of the basis. The Minimum Spanning Tree Cycle Intersection problem consists in finding a spanning tree such that the intersection number is minimum. This problem is relevant in order to integrate discrete differential forms. In this paper, we present two lower bounds of the intersection number of an arbitrary connected graph $G=(V,E)$. In the first part, we prove the following statement: $$\frac{1}{2}\left(\frac{\nu^2}{n-1} - \nu\right) \leq \cap(G),$$ where $n = |V|$ and $\nu$ is the \emph{cyclomatic number} of $G$. In the second part, based on some experimental results and a new observation, we conjecture the following improved tight lower bound: $$(n-1) \binom{q}{2} + q \ r\leq \cap(G),$$ where $2 \nu = q (n-1) + r$ is the integer division of $2 \nu$ and $n-1$. This is the first result in a general context, that is for an arbitrary connected graph.
2303.10590
Minh Tran
Yufeng Yin, Minh Tran, Di Chang, Xinrui Wang, Mohammad Soleymani
Multi-modal Facial Action Unit Detection with Large Pre-trained Models for the 5th Competition on Affective Behavior Analysis in-the-wild
8 pages, 7 figures, 5 tables
null
null
null
cs.CV
http://creativecommons.org/licenses/by/4.0/
Facial action unit detection has emerged as an important task within facial expression analysis, aimed at detecting specific pre-defined, objective facial expressions, such as lip tightening and cheek raising. This paper presents our submission to the Affective Behavior Analysis in-the-wild (ABAW) 2023 Competition for AU detection. We propose a multi-modal method for facial action unit detection with visual, acoustic, and lexical features extracted from the large pre-trained models. To provide high-quality details for visual feature extraction, we apply super-resolution and face alignment to the training data and show potential performance gain. Our approach achieves the F1 score of 52.3% on the official validation set of the 5th ABAW Challenge.
[ { "created": "Sun, 19 Mar 2023 07:18:14 GMT", "version": "v1" }, { "created": "Thu, 23 Mar 2023 00:35:40 GMT", "version": "v2" }, { "created": "Mon, 17 Apr 2023 20:17:55 GMT", "version": "v3" } ]
2023-04-19
[ [ "Yin", "Yufeng", "" ], [ "Tran", "Minh", "" ], [ "Chang", "Di", "" ], [ "Wang", "Xinrui", "" ], [ "Soleymani", "Mohammad", "" ] ]
Facial action unit detection has emerged as an important task within facial expression analysis, aimed at detecting specific pre-defined, objective facial expressions, such as lip tightening and cheek raising. This paper presents our submission to the Affective Behavior Analysis in-the-wild (ABAW) 2023 Competition for AU detection. We propose a multi-modal method for facial action unit detection with visual, acoustic, and lexical features extracted from the large pre-trained models. To provide high-quality details for visual feature extraction, we apply super-resolution and face alignment to the training data and show potential performance gain. Our approach achieves the F1 score of 52.3% on the official validation set of the 5th ABAW Challenge.
2310.02807
Zijie Geng
Zijie Geng, Xijun Li, Jie Wang, Xiao Li, Yongdong Zhang, Feng Wu
A Deep Instance Generative Framework for MILP Solvers Under Limited Data Availability
null
null
null
null
cs.LG
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
In the past few years, there has been an explosive surge in the use of machine learning (ML) techniques to address combinatorial optimization (CO) problems, especially mixed-integer linear programs (MILPs). Despite the achievements, the limited availability of real-world instances often leads to sub-optimal decisions and biased solver assessments, which motivates a suite of synthetic MILP instance generation techniques. However, existing methods either rely heavily on expert-designed formulations or struggle to capture the rich features of real-world instances. To tackle this problem, we propose G2MILP, the first deep generative framework for MILP instances. Specifically, G2MILP represents MILP instances as bipartite graphs, and applies a masked variational autoencoder to iteratively corrupt and replace parts of the original graphs to generate new ones. The appealing feature of G2MILP is that it can learn to generate novel and realistic MILP instances without prior expert-designed formulations, while preserving the structures and computational hardness of real-world datasets, simultaneously. Thus the generated instances can facilitate downstream tasks for enhancing MILP solvers under limited data availability. We design a suite of benchmarks to evaluate the quality of the generated MILP instances. Experiments demonstrate that our method can produce instances that closely resemble real-world datasets in terms of both structures and computational hardness. The deliverables are released at https://miralab-ustc.github.io/L2O-G2MILP.
[ { "created": "Wed, 4 Oct 2023 13:34:34 GMT", "version": "v1" }, { "created": "Sat, 28 Oct 2023 12:10:46 GMT", "version": "v2" }, { "created": "Mon, 11 Mar 2024 10:51:14 GMT", "version": "v3" } ]
2024-03-12
[ [ "Geng", "Zijie", "" ], [ "Li", "Xijun", "" ], [ "Wang", "Jie", "" ], [ "Li", "Xiao", "" ], [ "Zhang", "Yongdong", "" ], [ "Wu", "Feng", "" ] ]
In the past few years, there has been an explosive surge in the use of machine learning (ML) techniques to address combinatorial optimization (CO) problems, especially mixed-integer linear programs (MILPs). Despite the achievements, the limited availability of real-world instances often leads to sub-optimal decisions and biased solver assessments, which motivates a suite of synthetic MILP instance generation techniques. However, existing methods either rely heavily on expert-designed formulations or struggle to capture the rich features of real-world instances. To tackle this problem, we propose G2MILP, the first deep generative framework for MILP instances. Specifically, G2MILP represents MILP instances as bipartite graphs, and applies a masked variational autoencoder to iteratively corrupt and replace parts of the original graphs to generate new ones. The appealing feature of G2MILP is that it can learn to generate novel and realistic MILP instances without prior expert-designed formulations, while preserving the structures and computational hardness of real-world datasets, simultaneously. Thus the generated instances can facilitate downstream tasks for enhancing MILP solvers under limited data availability. We design a suite of benchmarks to evaluate the quality of the generated MILP instances. Experiments demonstrate that our method can produce instances that closely resemble real-world datasets in terms of both structures and computational hardness. The deliverables are released at https://miralab-ustc.github.io/L2O-G2MILP.
2102.00287
Eva Vanmassenhove
Eva Vanmassenhove, Dimitar Shterionov, Matthew Gwilliam
Machine Translationese: Effects of Algorithmic Bias on Linguistic Complexity in Machine Translation
null
null
null
null
cs.CL cs.AI cs.CY
http://creativecommons.org/licenses/by/4.0/
Recent studies in the field of Machine Translation (MT) and Natural Language Processing (NLP) have shown that existing models amplify biases observed in the training data. The amplification of biases in language technology has mainly been examined with respect to specific phenomena, such as gender bias. In this work, we go beyond the study of gender in MT and investigate how bias amplification might affect language in a broader sense. We hypothesize that the 'algorithmic bias', i.e. an exacerbation of frequently observed patterns in combination with a loss of less frequent ones, not only exacerbates societal biases present in current datasets but could also lead to an artificially impoverished language: 'machine translationese'. We assess the linguistic richness (on a lexical and morphological level) of translations created by different data-driven MT paradigms - phrase-based statistical (PB-SMT) and neural MT (NMT). Our experiments show that there is a loss of lexical and morphological richness in the translations produced by all investigated MT paradigms for two language pairs (EN<=>FR and EN<=>ES).
[ { "created": "Sat, 30 Jan 2021 18:49:11 GMT", "version": "v1" } ]
2021-02-02
[ [ "Vanmassenhove", "Eva", "" ], [ "Shterionov", "Dimitar", "" ], [ "Gwilliam", "Matthew", "" ] ]
Recent studies in the field of Machine Translation (MT) and Natural Language Processing (NLP) have shown that existing models amplify biases observed in the training data. The amplification of biases in language technology has mainly been examined with respect to specific phenomena, such as gender bias. In this work, we go beyond the study of gender in MT and investigate how bias amplification might affect language in a broader sense. We hypothesize that the 'algorithmic bias', i.e. an exacerbation of frequently observed patterns in combination with a loss of less frequent ones, not only exacerbates societal biases present in current datasets but could also lead to an artificially impoverished language: 'machine translationese'. We assess the linguistic richness (on a lexical and morphological level) of translations created by different data-driven MT paradigms - phrase-based statistical (PB-SMT) and neural MT (NMT). Our experiments show that there is a loss of lexical and morphological richness in the translations produced by all investigated MT paradigms for two language pairs (EN<=>FR and EN<=>ES).
0805.0873
EDA Publishing Association
Hela Boussetta (TIMA), S. Basrour (TIMA), M. Marzencki (TIMA)
Top-Down Behavioral Modeling Methodology of a Piezoelectric Microgenerator For Integrated Power Harvesting Systems
Submitted on behalf of EDA Publishing Association (http://irevues.inist.fr/handle/2042/16838)
Dans Symposium on Design, Test, Integration and Packaging of MEMS/MOEMS - DTIP 2008, Nice : France (2008)
null
null
cs.OH
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
In this study, we developed a top/down methodology for behavioral and structural modeling of multi-domain microsystems. Then, we validated this methodology through a study case : a piezoelectric microgenerator. We also proved the effectiveness of VHDL-AMS language not only for modeling in behavioral and structural levels but also in writing physical models that can predict the experimental results. Finally, we validated these models by presenting and discussing simulations results.
[ { "created": "Wed, 7 May 2008 09:00:16 GMT", "version": "v1" } ]
2008-12-18
[ [ "Boussetta", "Hela", "", "TIMA" ], [ "Basrour", "S.", "", "TIMA" ], [ "Marzencki", "M.", "", "TIMA" ] ]
In this study, we developed a top/down methodology for behavioral and structural modeling of multi-domain microsystems. Then, we validated this methodology through a study case : a piezoelectric microgenerator. We also proved the effectiveness of VHDL-AMS language not only for modeling in behavioral and structural levels but also in writing physical models that can predict the experimental results. Finally, we validated these models by presenting and discussing simulations results.
2405.00885
Huai-An Su
Huai-an Su, Jiaxiang Geng, Liang Li, Xiaoqi Qin, Yanzhao Hou, Xin Fu and Miao Pan
WHALE-FL: Wireless and Heterogeneity Aware Latency Efficient Federated Learning over Mobile Devices via Adaptive Subnetwork Scheduling
null
null
null
null
cs.LG cs.NI eess.IV
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
As a popular distributed learning paradigm, federated learning (FL) over mobile devices fosters numerous applications, while their practical deployment is hindered by participating devices' computing and communication heterogeneity. Some pioneering research efforts proposed to extract subnetworks from the global model, and assign as large a subnetwork as possible to the device for local training based on its full computing and communications capacity. Although such fixed size subnetwork assignment enables FL training over heterogeneous mobile devices, it is unaware of (i) the dynamic changes of devices' communication and computing conditions and (ii) FL training progress and its dynamic requirements of local training contributions, both of which may cause very long FL training delay. Motivated by those dynamics, in this paper, we develop a wireless and heterogeneity aware latency efficient FL (WHALE-FL) approach to accelerate FL training through adaptive subnetwork scheduling. Instead of sticking to the fixed size subnetwork, WHALE-FL introduces a novel subnetwork selection utility function to capture device and FL training dynamics, and guides the mobile device to adaptively select the subnetwork size for local training based on (a) its computing and communication capacity, (b) its dynamic computing and/or communication conditions, and (c) FL training status and its corresponding requirements for local training contributions. Our evaluation shows that, compared with peer designs, WHALE-FL effectively accelerates FL training without sacrificing learning accuracy.
[ { "created": "Wed, 1 May 2024 22:01:40 GMT", "version": "v1" } ]
2024-05-03
[ [ "Su", "Huai-an", "" ], [ "Geng", "Jiaxiang", "" ], [ "Li", "Liang", "" ], [ "Qin", "Xiaoqi", "" ], [ "Hou", "Yanzhao", "" ], [ "Fu", "Xin", "" ], [ "Pan", "Miao", "" ] ]
As a popular distributed learning paradigm, federated learning (FL) over mobile devices fosters numerous applications, while their practical deployment is hindered by participating devices' computing and communication heterogeneity. Some pioneering research efforts proposed to extract subnetworks from the global model, and assign as large a subnetwork as possible to the device for local training based on its full computing and communications capacity. Although such fixed size subnetwork assignment enables FL training over heterogeneous mobile devices, it is unaware of (i) the dynamic changes of devices' communication and computing conditions and (ii) FL training progress and its dynamic requirements of local training contributions, both of which may cause very long FL training delay. Motivated by those dynamics, in this paper, we develop a wireless and heterogeneity aware latency efficient FL (WHALE-FL) approach to accelerate FL training through adaptive subnetwork scheduling. Instead of sticking to the fixed size subnetwork, WHALE-FL introduces a novel subnetwork selection utility function to capture device and FL training dynamics, and guides the mobile device to adaptively select the subnetwork size for local training based on (a) its computing and communication capacity, (b) its dynamic computing and/or communication conditions, and (c) FL training status and its corresponding requirements for local training contributions. Our evaluation shows that, compared with peer designs, WHALE-FL effectively accelerates FL training without sacrificing learning accuracy.
2402.15145
Hongxun Wu
Xin Lyu, Hongxun Wu, Junzhao Yang
The Cost of Parallelizing Boosting
appeared in SODA 2024
null
null
null
cs.LG cs.DS
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
We study the cost of parallelizing weak-to-strong boosting algorithms for learning, following the recent work of Karbasi and Larsen. Our main results are two-fold: - First, we prove a tight lower bound, showing that even "slight" parallelization of boosting requires an exponential blow-up in the complexity of training. Specifically, let $\gamma$ be the weak learner's advantage over random guessing. The famous \textsc{AdaBoost} algorithm produces an accurate hypothesis by interacting with the weak learner for $\tilde{O}(1 / \gamma^2)$ rounds where each round runs in polynomial time. Karbasi and Larsen showed that "significant" parallelization must incur exponential blow-up: Any boosting algorithm either interacts with the weak learner for $\Omega(1 / \gamma)$ rounds or incurs an $\exp(d / \gamma)$ blow-up in the complexity of training, where $d$ is the VC dimension of the hypothesis class. We close the gap by showing that any boosting algorithm either has $\Omega(1 / \gamma^2)$ rounds of interaction or incurs a smaller exponential blow-up of $\exp(d)$. -Complementing our lower bound, we show that there exists a boosting algorithm using $\tilde{O}(1/(t \gamma^2))$ rounds, and only suffer a blow-up of $\exp(d \cdot t^2)$. Plugging in $t = \omega(1)$, this shows that the smaller blow-up in our lower bound is tight. More interestingly, this provides the first trade-off between the parallelism and the total work required for boosting.
[ { "created": "Fri, 23 Feb 2024 07:03:52 GMT", "version": "v1" } ]
2024-02-26
[ [ "Lyu", "Xin", "" ], [ "Wu", "Hongxun", "" ], [ "Yang", "Junzhao", "" ] ]
We study the cost of parallelizing weak-to-strong boosting algorithms for learning, following the recent work of Karbasi and Larsen. Our main results are two-fold: - First, we prove a tight lower bound, showing that even "slight" parallelization of boosting requires an exponential blow-up in the complexity of training. Specifically, let $\gamma$ be the weak learner's advantage over random guessing. The famous \textsc{AdaBoost} algorithm produces an accurate hypothesis by interacting with the weak learner for $\tilde{O}(1 / \gamma^2)$ rounds where each round runs in polynomial time. Karbasi and Larsen showed that "significant" parallelization must incur exponential blow-up: Any boosting algorithm either interacts with the weak learner for $\Omega(1 / \gamma)$ rounds or incurs an $\exp(d / \gamma)$ blow-up in the complexity of training, where $d$ is the VC dimension of the hypothesis class. We close the gap by showing that any boosting algorithm either has $\Omega(1 / \gamma^2)$ rounds of interaction or incurs a smaller exponential blow-up of $\exp(d)$. -Complementing our lower bound, we show that there exists a boosting algorithm using $\tilde{O}(1/(t \gamma^2))$ rounds, and only suffer a blow-up of $\exp(d \cdot t^2)$. Plugging in $t = \omega(1)$, this shows that the smaller blow-up in our lower bound is tight. More interestingly, this provides the first trade-off between the parallelism and the total work required for boosting.
2404.00728
Luis Morales-Navarro
Luis Morales-Navarro and Yasmin B. Kafai
Investigating Youths' Everyday Understanding of Machine Learning Applications: a Knowledge-in-Pieces Perspective
accepted for publication at Proceedings of the International Conference of the Learning Sciences 2024
null
null
null
cs.CY
http://creativecommons.org/licenses/by-nc-nd/4.0/
Despite recent calls for including artificial intelligence (AI) literacy in K-12 education, not enough attention has been paid to studying youths' everyday knowledge about machine learning (ML). Most research has examined how youths attribute intelligence to AI/ML systems. Other studies have centered on youths' theories and hypotheses about ML highlighting their misconceptions and how these may hinder learning. However, research on conceptual change shows that youths may not have coherent theories about scientific phenomena and instead have knowledge pieces that can be productive for formal learning. We investigate teens' everyday understanding of ML through a knowledge-in-pieces perspective. Our analyses reveal that youths showed some understanding that ML applications learn from training data and that applications recognize patterns in input data and depending on these provide different outputs. We discuss how these findings expand our knowledge base and implications for the design of tools and activities to introduce youths to ML.
[ { "created": "Sun, 31 Mar 2024 16:11:33 GMT", "version": "v1" } ]
2024-04-02
[ [ "Morales-Navarro", "Luis", "" ], [ "Kafai", "Yasmin B.", "" ] ]
Despite recent calls for including artificial intelligence (AI) literacy in K-12 education, not enough attention has been paid to studying youths' everyday knowledge about machine learning (ML). Most research has examined how youths attribute intelligence to AI/ML systems. Other studies have centered on youths' theories and hypotheses about ML highlighting their misconceptions and how these may hinder learning. However, research on conceptual change shows that youths may not have coherent theories about scientific phenomena and instead have knowledge pieces that can be productive for formal learning. We investigate teens' everyday understanding of ML through a knowledge-in-pieces perspective. Our analyses reveal that youths showed some understanding that ML applications learn from training data and that applications recognize patterns in input data and depending on these provide different outputs. We discuss how these findings expand our knowledge base and implications for the design of tools and activities to introduce youths to ML.
2103.00497
Aryan Asadian
Aryan Asadian, Amirali Salehi-Abari
Distilling Knowledge via Intermediate Classifiers
8 pages, 2 figures
null
null
null
cs.LG cs.AI cs.CV
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
The crux of knowledge distillation is to effectively train a resource-limited student model with the guide of a pre-trained larger teacher model. However, when there is a large difference between the model complexities of teacher and student (i.e., capacity gap), knowledge distillation loses its strength in transferring knowledge from the teacher to the student, thus training a weaker student. To mitigate the impact of the capacity gap, we introduce knowledge distillation via intermediate heads. By extending the intermediate layers of the teacher (at various depths) with classifier heads, we cheaply acquire a cohort of heterogeneous pre-trained teachers. The intermediate classifier heads can all together be efficiently learned while freezing the backbone of the pre-trained teacher. The cohort of teachers (including the original teacher) co-teach the student simultaneously. Our experiments on various teacher-student pairs and datasets have demonstrated that the proposed approach outperforms the canonical knowledge distillation approach and its extensions.
[ { "created": "Sun, 28 Feb 2021 12:52:52 GMT", "version": "v1" }, { "created": "Mon, 31 May 2021 13:20:57 GMT", "version": "v2" } ]
2021-06-01
[ [ "Asadian", "Aryan", "" ], [ "Salehi-Abari", "Amirali", "" ] ]
The crux of knowledge distillation is to effectively train a resource-limited student model with the guide of a pre-trained larger teacher model. However, when there is a large difference between the model complexities of teacher and student (i.e., capacity gap), knowledge distillation loses its strength in transferring knowledge from the teacher to the student, thus training a weaker student. To mitigate the impact of the capacity gap, we introduce knowledge distillation via intermediate heads. By extending the intermediate layers of the teacher (at various depths) with classifier heads, we cheaply acquire a cohort of heterogeneous pre-trained teachers. The intermediate classifier heads can all together be efficiently learned while freezing the backbone of the pre-trained teacher. The cohort of teachers (including the original teacher) co-teach the student simultaneously. Our experiments on various teacher-student pairs and datasets have demonstrated that the proposed approach outperforms the canonical knowledge distillation approach and its extensions.
1708.06805
Jordi Levy
Carlos Ans\'otegui, Maria Luisa Bonet, Jordi Levy
Scale-Free Random SAT Instances
null
Algorithms 15(6): 219 (2022)
10.3390/a15060219
null
cs.CC math.CO math.PR
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
We focus on the random generation of SAT instances that have properties similar to real-world instances. It is known that many industrial instances, even with a great number of variables, can be solved by a clever solver in a reasonable amount of time. This is not possible, in general, with classical randomly generated instances. We provide a different generation model of SAT instances, called \emph{scale-free random SAT instances}. It is based on the use of a non-uniform probability distribution $P(i)\sim i^{-\beta}$ to select variable $i$, where $\beta$ is a parameter of the model. This results into formulas where the number of occurrences $k$ of variables follows a power-law distribution $P(k)\sim k^{-\delta}$ where $\delta = 1 + 1/\beta$. This property has been observed in most real-world SAT instances. For $\beta=0$, our model extends classical random SAT instances. We prove the existence of a SAT-UNSAT phase transition phenomenon for scale-free random 2-SAT instances with $\beta<1/2$ when the clause/variable ratio is $m/n=\frac{1-2\beta}{(1-\beta)^2}$. We also prove that scale-free random k-SAT instances are unsatisfiable with high probability when the number of clauses exceeds $\omega(n^{(1-\beta)k})$. %This implies that the SAT/UNSAT phase transition phenomena vanishes when $\beta>1-1/k$, and formulas are unsatisfiable due to a small core of clauses. The proof of this result suggests that, when $\beta>1-1/k$, the unsatisfiability of most formulas may be due to small cores of clauses. Finally, we show how this model will allow us to generate random instances similar to industrial instances, of interest for testing purposes.
[ { "created": "Wed, 12 Jul 2017 19:21:19 GMT", "version": "v1" }, { "created": "Tue, 9 Apr 2019 16:49:29 GMT", "version": "v2" }, { "created": "Wed, 17 Jul 2019 21:49:35 GMT", "version": "v3" } ]
2023-03-14
[ [ "Ansótegui", "Carlos", "" ], [ "Bonet", "Maria Luisa", "" ], [ "Levy", "Jordi", "" ] ]
We focus on the random generation of SAT instances that have properties similar to real-world instances. It is known that many industrial instances, even with a great number of variables, can be solved by a clever solver in a reasonable amount of time. This is not possible, in general, with classical randomly generated instances. We provide a different generation model of SAT instances, called \emph{scale-free random SAT instances}. It is based on the use of a non-uniform probability distribution $P(i)\sim i^{-\beta}$ to select variable $i$, where $\beta$ is a parameter of the model. This results into formulas where the number of occurrences $k$ of variables follows a power-law distribution $P(k)\sim k^{-\delta}$ where $\delta = 1 + 1/\beta$. This property has been observed in most real-world SAT instances. For $\beta=0$, our model extends classical random SAT instances. We prove the existence of a SAT-UNSAT phase transition phenomenon for scale-free random 2-SAT instances with $\beta<1/2$ when the clause/variable ratio is $m/n=\frac{1-2\beta}{(1-\beta)^2}$. We also prove that scale-free random k-SAT instances are unsatisfiable with high probability when the number of clauses exceeds $\omega(n^{(1-\beta)k})$. %This implies that the SAT/UNSAT phase transition phenomena vanishes when $\beta>1-1/k$, and formulas are unsatisfiable due to a small core of clauses. The proof of this result suggests that, when $\beta>1-1/k$, the unsatisfiability of most formulas may be due to small cores of clauses. Finally, we show how this model will allow us to generate random instances similar to industrial instances, of interest for testing purposes.
1510.00921
Chunhua Shen
Lingqiao Liu, Chunhua Shen, Anton van den Hengel
Cross-convolutional-layer Pooling for Image Recognition
Fixed typos. Journal extension of arXiv:1411.7466. Accepted to IEEE Transactions on Pattern Analysis and Machine Intelligence
null
null
null
cs.CV
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
Recent studies have shown that a Deep Convolutional Neural Network (DCNN) pretrained on a large image dataset can be used as a universal image descriptor, and that doing so leads to impressive performance for a variety of image classification tasks. Most of these studies adopt activations from a single DCNN layer, usually the fully-connected layer, as the image representation. In this paper, we proposed a novel way to extract image representations from two consecutive convolutional layers: one layer is utilized for local feature extraction and the other serves as guidance to pool the extracted features. By taking different viewpoints of convolutional layers, we further develop two schemes to realize this idea. The first one directly uses convolutional layers from a DCNN. The second one applies the pretrained CNN on densely sampled image regions and treats the fully-connected activations of each image region as convolutional feature activations. We then train another convolutional layer on top of that as the pooling-guidance convolutional layer. By applying our method to three popular visual classification tasks, we find our first scheme tends to perform better on the applications which need strong discrimination on subtle object patterns within small regions while the latter excels in the cases that require discrimination on category-level patterns. Overall, the proposed method achieves superior performance over existing ways of extracting image representations from a DCNN.
[ { "created": "Sun, 4 Oct 2015 10:27:36 GMT", "version": "v1" }, { "created": "Thu, 9 Jun 2016 07:37:09 GMT", "version": "v2" }, { "created": "Sun, 23 Oct 2016 05:48:16 GMT", "version": "v3" }, { "created": "Wed, 7 Dec 2016 00:00:42 GMT", "version": "v4" }, { "created": "Thu, 8 Dec 2016 01:31:05 GMT", "version": "v5" }, { "created": "Thu, 22 Dec 2016 04:43:19 GMT", "version": "v6" } ]
2016-12-23
[ [ "Liu", "Lingqiao", "" ], [ "Shen", "Chunhua", "" ], [ "Hengel", "Anton van den", "" ] ]
Recent studies have shown that a Deep Convolutional Neural Network (DCNN) pretrained on a large image dataset can be used as a universal image descriptor, and that doing so leads to impressive performance for a variety of image classification tasks. Most of these studies adopt activations from a single DCNN layer, usually the fully-connected layer, as the image representation. In this paper, we proposed a novel way to extract image representations from two consecutive convolutional layers: one layer is utilized for local feature extraction and the other serves as guidance to pool the extracted features. By taking different viewpoints of convolutional layers, we further develop two schemes to realize this idea. The first one directly uses convolutional layers from a DCNN. The second one applies the pretrained CNN on densely sampled image regions and treats the fully-connected activations of each image region as convolutional feature activations. We then train another convolutional layer on top of that as the pooling-guidance convolutional layer. By applying our method to three popular visual classification tasks, we find our first scheme tends to perform better on the applications which need strong discrimination on subtle object patterns within small regions while the latter excels in the cases that require discrimination on category-level patterns. Overall, the proposed method achieves superior performance over existing ways of extracting image representations from a DCNN.
2403.15122
Sacha-\'Elie Ayoun
Sacha-\'Elie Ayoun, Xavier Denis, Petar Maksimovi\'c, Philippa Gardner
A hybrid approach to semi-automated Rust verification
22 pages, 8 figures, preprint
null
null
null
cs.PL
http://creativecommons.org/licenses/by/4.0/
While recent years have been witness to a large body of work on efficient and automated verification of safe Rust code, enabled by the rich guarantees of the Rust type system, much less progress has been made on reasoning about unsafe code due to its unique complexities. We propose a hybrid approach to end-to-end Rust verification in which powerful automated verification of safe Rust is combined with targeted semi-automated verification of unsafe~Rust. To this end, we present Gillian-Rust, a proof-of-concept semi-automated verification tool that is able to reason about type safety and functional correctness of unsafe~code. Built on top of the Gillian parametric compositional verification platform, Gillian-Rust automates a rich separation logic for real-world Rust, embedding the lifetime logic of RustBelt and the parametric propheciees of RustHornBelt. Using the unique extensibility of Gillian, our novel encoding of these features is fine-tuned to maximise automation and exposes a user-friendly API, allowing for low-effort verification of unsafe code. We link Gillian-Rust with Creusot, a state-of-the-art verifier for safe Rust, by providing a systematic encoding of unsafe code specifications that Creusot may use but not verify, demonstrating the feasibility of our hybrid~approach.
[ { "created": "Fri, 22 Mar 2024 11:24:31 GMT", "version": "v1" } ]
2024-03-25
[ [ "Ayoun", "Sacha-Élie", "" ], [ "Denis", "Xavier", "" ], [ "Maksimović", "Petar", "" ], [ "Gardner", "Philippa", "" ] ]
While recent years have been witness to a large body of work on efficient and automated verification of safe Rust code, enabled by the rich guarantees of the Rust type system, much less progress has been made on reasoning about unsafe code due to its unique complexities. We propose a hybrid approach to end-to-end Rust verification in which powerful automated verification of safe Rust is combined with targeted semi-automated verification of unsafe~Rust. To this end, we present Gillian-Rust, a proof-of-concept semi-automated verification tool that is able to reason about type safety and functional correctness of unsafe~code. Built on top of the Gillian parametric compositional verification platform, Gillian-Rust automates a rich separation logic for real-world Rust, embedding the lifetime logic of RustBelt and the parametric propheciees of RustHornBelt. Using the unique extensibility of Gillian, our novel encoding of these features is fine-tuned to maximise automation and exposes a user-friendly API, allowing for low-effort verification of unsafe code. We link Gillian-Rust with Creusot, a state-of-the-art verifier for safe Rust, by providing a systematic encoding of unsafe code specifications that Creusot may use but not verify, demonstrating the feasibility of our hybrid~approach.
0909.3648
Joel Ratsaby
Joel Ratsaby
Random scattering of bits by prediction
null
null
null
null
cs.AI cs.IT math.IT
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
We investigate a population of binary mistake sequences that result from learning with parametric models of different order. We obtain estimates of their error, algorithmic complexity and divergence from a purely random Bernoulli sequence. We study the relationship of these variables to the learner's information density parameter which is defined as the ratio between the lengths of the compressed to uncompressed files that contain the learner's decision rule. The results indicate that good learners have a low information density$\rho$ while bad learners have a high $\rho$. Bad learners generate mistake sequences that are atypically complex or diverge stochastically from a purely random Bernoulli sequence. Good learners generate typically complex sequences with low divergence from Bernoulli sequences and they include mistake sequences generated by the Bayes optimal predictor. Based on the static algorithmic interference model of \cite{Ratsaby_entropy} the learner here acts as a static structure which "scatters" the bits of an input sequence (to be predicted) in proportion to its information density $\rho$ thereby deforming its randomness characteristics.
[ { "created": "Sun, 20 Sep 2009 18:10:55 GMT", "version": "v1" }, { "created": "Wed, 13 Oct 2010 19:03:58 GMT", "version": "v2" } ]
2010-10-14
[ [ "Ratsaby", "Joel", "" ] ]
We investigate a population of binary mistake sequences that result from learning with parametric models of different order. We obtain estimates of their error, algorithmic complexity and divergence from a purely random Bernoulli sequence. We study the relationship of these variables to the learner's information density parameter which is defined as the ratio between the lengths of the compressed to uncompressed files that contain the learner's decision rule. The results indicate that good learners have a low information density$\rho$ while bad learners have a high $\rho$. Bad learners generate mistake sequences that are atypically complex or diverge stochastically from a purely random Bernoulli sequence. Good learners generate typically complex sequences with low divergence from Bernoulli sequences and they include mistake sequences generated by the Bayes optimal predictor. Based on the static algorithmic interference model of \cite{Ratsaby_entropy} the learner here acts as a static structure which "scatters" the bits of an input sequence (to be predicted) in proportion to its information density $\rho$ thereby deforming its randomness characteristics.
cs/0611048
Richard Mayr
Parosh Abdulla, Pritha Mahata, Richard Mayr
Dense-Timed Petri Nets: Checking Zenoness, Token liveness and Boundedness
61 pages, 18 figures
Logical Methods in Computer Science, Volume 3, Issue 1 (February 7, 2007) lmcs:2223
10.2168/LMCS-3(1:1)2007
null
cs.LO
null
We consider Dense-Timed Petri Nets (TPN), an extension of Petri nets in which each token is equipped with a real-valued clock and where the semantics is lazy (i.e., enabled transitions need not fire; time can pass and disable transitions). We consider the following verification problems for TPNs. (i) Zenoness: whether there exists a zeno-computation from a given marking, i.e., an infinite computation which takes only a finite amount of time. We show decidability of zenoness for TPNs, thus solving an open problem from [Escrig et al.]. Furthermore, the related question if there exist arbitrarily fast computations from a given marking is also decidable. On the other hand, universal zenoness, i.e., the question if all infinite computations from a given marking are zeno, is undecidable. (ii) Token liveness: whether a token is alive in a marking, i.e., whether there is a computation from the marking which eventually consumes the token. We show decidability of the problem by reducing it to the coverability problem, which is decidable for TPNs. (iii) Boundedness: whether the size of the reachable markings is bounded. We consider two versions of the problem; namely semantic boundedness where only live tokens are taken into consideration in the markings, and syntactic boundedness where also dead tokens are considered. We show undecidability of semantic boundedness, while we prove that syntactic boundedness is decidable through an extension of the Karp-Miller algorithm.
[ { "created": "Sat, 11 Nov 2006 00:08:46 GMT", "version": "v1" }, { "created": "Tue, 23 Jan 2007 13:33:21 GMT", "version": "v2" } ]
2017-01-11
[ [ "Abdulla", "Parosh", "" ], [ "Mahata", "Pritha", "" ], [ "Mayr", "Richard", "" ] ]
We consider Dense-Timed Petri Nets (TPN), an extension of Petri nets in which each token is equipped with a real-valued clock and where the semantics is lazy (i.e., enabled transitions need not fire; time can pass and disable transitions). We consider the following verification problems for TPNs. (i) Zenoness: whether there exists a zeno-computation from a given marking, i.e., an infinite computation which takes only a finite amount of time. We show decidability of zenoness for TPNs, thus solving an open problem from [Escrig et al.]. Furthermore, the related question if there exist arbitrarily fast computations from a given marking is also decidable. On the other hand, universal zenoness, i.e., the question if all infinite computations from a given marking are zeno, is undecidable. (ii) Token liveness: whether a token is alive in a marking, i.e., whether there is a computation from the marking which eventually consumes the token. We show decidability of the problem by reducing it to the coverability problem, which is decidable for TPNs. (iii) Boundedness: whether the size of the reachable markings is bounded. We consider two versions of the problem; namely semantic boundedness where only live tokens are taken into consideration in the markings, and syntactic boundedness where also dead tokens are considered. We show undecidability of semantic boundedness, while we prove that syntactic boundedness is decidable through an extension of the Karp-Miller algorithm.
1202.0533
Mark Wilde
Saikat Guha and Mark M. Wilde
Polar coding to achieve the Holevo capacity of a pure-loss optical channel
5 pages, submission to the 2012 International Symposium on Information Theory (ISIT 2012), Boston, MA, USA; v2 accepted to ISIT 2012
Proceedings of the 2012 IEEE International Symposium on Information Theory (ISIT 2012), pages 546-550, Cambridge, MA, USA
10.1109/ISIT.2012.6284250
null
cs.IT math.IT quant-ph
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
In the low-energy high-energy-efficiency regime of classical optical communications---relevant to deep-space optical channels---there is a big gap between reliable communication rates achievable via conventional optical receivers and the ultimate (Holevo) capacity. Achieving the Holevo capacity requires not only optimal codes but also receivers that make collective measurements on long (modulated) codeword waveforms, and it is impossible to implement these collective measurements via symbol-by-symbol detection along with classical postprocessing. Here, we apply our recent results on the classical-quantum polar code---the first near-explicit, linear, symmetric-Holevo-rate achieving code---to the lossy optical channel, and we show that it almost closes the entire gap to the Holevo capacity in the low photon number regime. In contrast, Arikan's original polar codes, applied to the DMC induced by the physical optical channel paired with any conceivable structured optical receiver (including optical homodyne, heterodyne, or direct-detection) fails to achieve the ultimate Holevo limit to channel capacity. However, our polar code construction (which uses the quantum fidelity as a channel parameter rather than the classical Bhattacharyya quantity to choose the "good channels" in the polar-code construction), paired with a quantum successive-cancellation receiver---which involves a sequence of collective non-destructive binary projective measurements on the joint quantum state of the received codeword waveform---can attain the Holevo limit, and can hence in principle achieve higher rates than Arikan's polar code and decoder directly applied to the optical channel. However, even a theoretical recipe for construction of an optical realization of the quantum successive-cancellation receiver remains an open question.
[ { "created": "Thu, 2 Feb 2012 20:01:30 GMT", "version": "v1" }, { "created": "Tue, 22 May 2012 12:22:05 GMT", "version": "v2" } ]
2012-09-04
[ [ "Guha", "Saikat", "" ], [ "Wilde", "Mark M.", "" ] ]
In the low-energy high-energy-efficiency regime of classical optical communications---relevant to deep-space optical channels---there is a big gap between reliable communication rates achievable via conventional optical receivers and the ultimate (Holevo) capacity. Achieving the Holevo capacity requires not only optimal codes but also receivers that make collective measurements on long (modulated) codeword waveforms, and it is impossible to implement these collective measurements via symbol-by-symbol detection along with classical postprocessing. Here, we apply our recent results on the classical-quantum polar code---the first near-explicit, linear, symmetric-Holevo-rate achieving code---to the lossy optical channel, and we show that it almost closes the entire gap to the Holevo capacity in the low photon number regime. In contrast, Arikan's original polar codes, applied to the DMC induced by the physical optical channel paired with any conceivable structured optical receiver (including optical homodyne, heterodyne, or direct-detection) fails to achieve the ultimate Holevo limit to channel capacity. However, our polar code construction (which uses the quantum fidelity as a channel parameter rather than the classical Bhattacharyya quantity to choose the "good channels" in the polar-code construction), paired with a quantum successive-cancellation receiver---which involves a sequence of collective non-destructive binary projective measurements on the joint quantum state of the received codeword waveform---can attain the Holevo limit, and can hence in principle achieve higher rates than Arikan's polar code and decoder directly applied to the optical channel. However, even a theoretical recipe for construction of an optical realization of the quantum successive-cancellation receiver remains an open question.
1910.10307
Vahdat Abdelzad
Vahdat Abdelzad, Krzysztof Czarnecki, Rick Salay, Taylor Denounden, Sachin Vernekar, Buu Phan
Detecting Out-of-Distribution Inputs in Deep Neural Networks Using an Early-Layer Output
15 pages, 8 figures
null
null
null
cs.LG stat.ML
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
Deep neural networks achieve superior performance in challenging tasks such as image classification. However, deep classifiers tend to incorrectly classify out-of-distribution (OOD) inputs, which are inputs that do not belong to the classifier training distribution. Several approaches have been proposed to detect OOD inputs, but the detection task is still an ongoing challenge. In this paper, we propose a new OOD detection approach that can be easily applied to an existing classifier and does not need to have access to OOD samples. The detector is a one-class classifier trained on the output of an early layer of the original classifier fed with its original training set. We apply our approach to several low- and high-dimensional datasets and compare it to the state-of-the-art detection approaches. Our approach achieves substantially better results over multiple metrics.
[ { "created": "Wed, 23 Oct 2019 01:27:48 GMT", "version": "v1" } ]
2019-10-24
[ [ "Abdelzad", "Vahdat", "" ], [ "Czarnecki", "Krzysztof", "" ], [ "Salay", "Rick", "" ], [ "Denounden", "Taylor", "" ], [ "Vernekar", "Sachin", "" ], [ "Phan", "Buu", "" ] ]
Deep neural networks achieve superior performance in challenging tasks such as image classification. However, deep classifiers tend to incorrectly classify out-of-distribution (OOD) inputs, which are inputs that do not belong to the classifier training distribution. Several approaches have been proposed to detect OOD inputs, but the detection task is still an ongoing challenge. In this paper, we propose a new OOD detection approach that can be easily applied to an existing classifier and does not need to have access to OOD samples. The detector is a one-class classifier trained on the output of an early layer of the original classifier fed with its original training set. We apply our approach to several low- and high-dimensional datasets and compare it to the state-of-the-art detection approaches. Our approach achieves substantially better results over multiple metrics.
cs/0703072
Paul Fodor
Paul Fodor
Domain Directed Dialogs for Decision Processes
null
null
null
null
cs.OH
null
The search for a standardized optimum way to communicate using natural language dialog has involved a lot of research. However, due to the diversity of communication domains, we think that this is extremely difficult to achieve and different dialogue management techniques should be applied for different situations. Our work presents the basis of a communication mechanism that supports decision processes, is based on decision trees, and minimizes the number of steps (turn-takes) in the dialogue. The initial dialog workflow is automatically generated and the user's interaction with the system can also change the decision tree and create new dialog paths with optimized cost. The decision tree represents the chronological ordering of the actions (via the parent-child relationship) and uses an object frame to represent the information state (capturing the notion of context). This paper presents our framework, the formalism for interaction and dialogue, and an evaluation of the system compared to relevant dialog planning frameworks (i.e. finite state diagrams, frame-based, information state and planning-based dialogue systems).
[ { "created": "Thu, 15 Mar 2007 00:08:50 GMT", "version": "v1" }, { "created": "Thu, 22 Mar 2007 14:54:41 GMT", "version": "v2" }, { "created": "Tue, 27 Mar 2007 15:41:54 GMT", "version": "v3" } ]
2007-05-23
[ [ "Fodor", "Paul", "" ] ]
The search for a standardized optimum way to communicate using natural language dialog has involved a lot of research. However, due to the diversity of communication domains, we think that this is extremely difficult to achieve and different dialogue management techniques should be applied for different situations. Our work presents the basis of a communication mechanism that supports decision processes, is based on decision trees, and minimizes the number of steps (turn-takes) in the dialogue. The initial dialog workflow is automatically generated and the user's interaction with the system can also change the decision tree and create new dialog paths with optimized cost. The decision tree represents the chronological ordering of the actions (via the parent-child relationship) and uses an object frame to represent the information state (capturing the notion of context). This paper presents our framework, the formalism for interaction and dialogue, and an evaluation of the system compared to relevant dialog planning frameworks (i.e. finite state diagrams, frame-based, information state and planning-based dialogue systems).
1410.0265
Chao Li
Chao Li, Michael Hay, Gerome Miklau, Yue Wang
A Data- and Workload-Aware Algorithm for Range Queries Under Differential Privacy
VLDB 2014
null
null
null
cs.DB
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
We describe a new algorithm for answering a given set of range queries under $\epsilon$-differential privacy which often achieves substantially lower error than competing methods. Our algorithm satisfies differential privacy by adding noise that is adapted to the input data and to the given query set. We first privately learn a partitioning of the domain into buckets that suit the input data well. Then we privately estimate counts for each bucket, doing so in a manner well-suited for the given query set. Since the performance of the algorithm depends on the input database, we evaluate it on a wide range of real datasets, showing that we can achieve the benefits of data-dependence on both "easy" and "hard" databases.
[ { "created": "Wed, 1 Oct 2014 15:56:42 GMT", "version": "v1" } ]
2014-10-02
[ [ "Li", "Chao", "" ], [ "Hay", "Michael", "" ], [ "Miklau", "Gerome", "" ], [ "Wang", "Yue", "" ] ]
We describe a new algorithm for answering a given set of range queries under $\epsilon$-differential privacy which often achieves substantially lower error than competing methods. Our algorithm satisfies differential privacy by adding noise that is adapted to the input data and to the given query set. We first privately learn a partitioning of the domain into buckets that suit the input data well. Then we privately estimate counts for each bucket, doing so in a manner well-suited for the given query set. Since the performance of the algorithm depends on the input database, we evaluate it on a wide range of real datasets, showing that we can achieve the benefits of data-dependence on both "easy" and "hard" databases.
1806.06084
Ramik Sadana
Ramik Sadana, Meeshu Agnihotri, John Stasko
Touching Data: A Discoverability-based Evaluation of a Visualization Interface for Tablet Computers
10 pages, 3 figures, 7 tabels
null
null
null
cs.HC
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
While a number of touch-based visualization systems have appeared in recent years, relatively little work has been done to evaluate these systems. The prevailing methods compare these systems to desktop-class applications or utilize traditional training-based usability studies. We argue that existing studies, while useful, fail to address a key aspect of mobile application usage - initial impression and discoverability-driven usability. Over the past few years, we have developed a tablet-based visualization system, Tangere, for analyzing tabular data in a multiple coordinated view configuration. This article describes a discoverability-based user study of Tangere in which the system is compared to a commercially available visualization system for tablets - Tableau's Vizable. The study highlights aspects of each system's design that resonate with study participants, and we reflect upon those findings to identify design principles for future tablet-based data visualization systems.
[ { "created": "Fri, 15 Jun 2018 18:17:34 GMT", "version": "v1" } ]
2018-06-19
[ [ "Sadana", "Ramik", "" ], [ "Agnihotri", "Meeshu", "" ], [ "Stasko", "John", "" ] ]
While a number of touch-based visualization systems have appeared in recent years, relatively little work has been done to evaluate these systems. The prevailing methods compare these systems to desktop-class applications or utilize traditional training-based usability studies. We argue that existing studies, while useful, fail to address a key aspect of mobile application usage - initial impression and discoverability-driven usability. Over the past few years, we have developed a tablet-based visualization system, Tangere, for analyzing tabular data in a multiple coordinated view configuration. This article describes a discoverability-based user study of Tangere in which the system is compared to a commercially available visualization system for tablets - Tableau's Vizable. The study highlights aspects of each system's design that resonate with study participants, and we reflect upon those findings to identify design principles for future tablet-based data visualization systems.
1611.07804
Nikita Astrakhantsev
N. Astrakhantsev
ATR4S: Toolkit with State-of-the-art Automatic Terms Recognition Methods in Scala
null
null
null
null
cs.CL
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
Automatically recognized terminology is widely used for various domain-specific texts processing tasks, such as machine translation, information retrieval or sentiment analysis. However, there is still no agreement on which methods are best suited for particular settings and, moreover, there is no reliable comparison of already developed methods. We believe that one of the main reasons is the lack of state-of-the-art methods implementations, which are usually non-trivial to recreate. In order to address these issues, we present ATR4S, an open-source software written in Scala that comprises more than 15 methods for automatic terminology recognition (ATR) and implements the whole pipeline from text document preprocessing, to term candidates collection, term candidates scoring, and finally, term candidates ranking. It is highly scalable, modular and configurable tool with support of automatic caching. We also compare 10 state-of-the-art methods on 7 open datasets by average precision and processing time. Experimental comparison reveals that no single method demonstrates best average precision for all datasets and that other available tools for ATR do not contain the best methods.
[ { "created": "Wed, 23 Nov 2016 14:14:52 GMT", "version": "v1" } ]
2016-11-24
[ [ "Astrakhantsev", "N.", "" ] ]
Automatically recognized terminology is widely used for various domain-specific texts processing tasks, such as machine translation, information retrieval or sentiment analysis. However, there is still no agreement on which methods are best suited for particular settings and, moreover, there is no reliable comparison of already developed methods. We believe that one of the main reasons is the lack of state-of-the-art methods implementations, which are usually non-trivial to recreate. In order to address these issues, we present ATR4S, an open-source software written in Scala that comprises more than 15 methods for automatic terminology recognition (ATR) and implements the whole pipeline from text document preprocessing, to term candidates collection, term candidates scoring, and finally, term candidates ranking. It is highly scalable, modular and configurable tool with support of automatic caching. We also compare 10 state-of-the-art methods on 7 open datasets by average precision and processing time. Experimental comparison reveals that no single method demonstrates best average precision for all datasets and that other available tools for ATR do not contain the best methods.
2203.08037
Yang Yang
Yang Yang, Xibai Lou, Changhyun Choi
Interactive Robotic Grasping with Attribute-Guided Disambiguation
Accepted to the IEEE International Conference on Robotics and Automation (ICRA 2022). Project page: https://sites.google.com/umn.edu/attr-disam
null
null
null
cs.RO
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
Interactive robotic grasping using natural language is one of the most fundamental tasks in human-robot interaction. However, language can be a source of ambiguity, particularly when there are ambiguous visual or linguistic contents. This paper investigates the use of object attributes in disambiguation and develops an interactive grasping system capable of effectively resolving ambiguities via dialogues. Our approach first predicts target scores and attribute scores through vision-and-language grounding. To handle ambiguous objects and commands, we propose an attribute-guided formulation of the partially observable Markov decision process (Attr-POMDP) for disambiguation. The Attr-POMDP utilizes target and attribute scores as the observation model to calculate the expected return of an attribute-based (e.g., "what is the color of the target, red or green?") or a pointing-based (e.g., "do you mean this one?") question. Our disambiguation module runs in real time on a real robot, and the interactive grasping system achieves a 91.43\% selection accuracy in the real-robot experiments, outperforming several baselines by large margins.
[ { "created": "Tue, 15 Mar 2022 16:17:36 GMT", "version": "v1" } ]
2022-03-16
[ [ "Yang", "Yang", "" ], [ "Lou", "Xibai", "" ], [ "Choi", "Changhyun", "" ] ]
Interactive robotic grasping using natural language is one of the most fundamental tasks in human-robot interaction. However, language can be a source of ambiguity, particularly when there are ambiguous visual or linguistic contents. This paper investigates the use of object attributes in disambiguation and develops an interactive grasping system capable of effectively resolving ambiguities via dialogues. Our approach first predicts target scores and attribute scores through vision-and-language grounding. To handle ambiguous objects and commands, we propose an attribute-guided formulation of the partially observable Markov decision process (Attr-POMDP) for disambiguation. The Attr-POMDP utilizes target and attribute scores as the observation model to calculate the expected return of an attribute-based (e.g., "what is the color of the target, red or green?") or a pointing-based (e.g., "do you mean this one?") question. Our disambiguation module runs in real time on a real robot, and the interactive grasping system achieves a 91.43\% selection accuracy in the real-robot experiments, outperforming several baselines by large margins.
2402.15392
Filippo Lazzati
Filippo Lazzati, Mirco Mutti, Alberto Maria Metelli
Offline Inverse RL: New Solution Concepts and Provably Efficient Algorithms
International Conference on Machine Learning 41 (ICML 2024)
null
null
null
cs.LG
http://creativecommons.org/licenses/by/4.0/
Inverse reinforcement learning (IRL) aims to recover the reward function of an expert agent from demonstrations of behavior. It is well-known that the IRL problem is fundamentally ill-posed, i.e., many reward functions can explain the demonstrations. For this reason, IRL has been recently reframed in terms of estimating the feasible reward set (Metelli et al., 2021), thus, postponing the selection of a single reward. However, so far, the available formulations and algorithmic solutions have been proposed and analyzed mainly for the online setting, where the learner can interact with the environment and query the expert at will. This is clearly unrealistic in most practical applications, where the availability of an offline dataset is a much more common scenario. In this paper, we introduce a novel notion of feasible reward set capturing the opportunities and limitations of the offline setting and we analyze the complexity of its estimation. This requires the introduction an original learning framework that copes with the intrinsic difficulty of the setting, for which the data coverage is not under control. Then, we propose two computationally and statistically efficient algorithms, IRLO and PIRLO, for addressing the problem. In particular, the latter adopts a specific form of pessimism to enforce the novel desirable property of inclusion monotonicity of the delivered feasible set. With this work, we aim to provide a panorama of the challenges of the offline IRL problem and how they can be fruitfully addressed.
[ { "created": "Fri, 23 Feb 2024 15:49:46 GMT", "version": "v1" }, { "created": "Thu, 6 Jun 2024 06:49:52 GMT", "version": "v2" } ]
2024-06-07
[ [ "Lazzati", "Filippo", "" ], [ "Mutti", "Mirco", "" ], [ "Metelli", "Alberto Maria", "" ] ]
Inverse reinforcement learning (IRL) aims to recover the reward function of an expert agent from demonstrations of behavior. It is well-known that the IRL problem is fundamentally ill-posed, i.e., many reward functions can explain the demonstrations. For this reason, IRL has been recently reframed in terms of estimating the feasible reward set (Metelli et al., 2021), thus, postponing the selection of a single reward. However, so far, the available formulations and algorithmic solutions have been proposed and analyzed mainly for the online setting, where the learner can interact with the environment and query the expert at will. This is clearly unrealistic in most practical applications, where the availability of an offline dataset is a much more common scenario. In this paper, we introduce a novel notion of feasible reward set capturing the opportunities and limitations of the offline setting and we analyze the complexity of its estimation. This requires the introduction an original learning framework that copes with the intrinsic difficulty of the setting, for which the data coverage is not under control. Then, we propose two computationally and statistically efficient algorithms, IRLO and PIRLO, for addressing the problem. In particular, the latter adopts a specific form of pessimism to enforce the novel desirable property of inclusion monotonicity of the delivered feasible set. With this work, we aim to provide a panorama of the challenges of the offline IRL problem and how they can be fruitfully addressed.
2401.00781
Shuang Li
Shuang Li, Ziyuan Pu, Zhiyong Cui, Seunghyeon Lee, Xiucheng Guo, Dong Ngoduy
Inferring Heterogeneous Treatment Effects of Crashes on Highway Traffic: A Doubly Robust Causal Machine Learning Approach
38 pages, 13 figures, 8 tables
null
null
null
cs.LG stat.ML
http://creativecommons.org/licenses/by/4.0/
Highway traffic crashes exert a considerable impact on both transportation systems and the economy. In this context, accurate and dependable emergency responses are crucial for effective traffic management. However, the influence of crashes on traffic status varies across diverse factors and may be biased due to selection bias. Therefore, there arises a necessity to accurately estimate the heterogeneous causal effects of crashes, thereby providing essential insights to facilitate individual-level emergency decision-making. This paper proposes a novel causal machine learning framework to estimate the causal effect of different types of crashes on highway speed. The Neyman-Rubin Causal Model (RCM) is employed to formulate this problem from a causal perspective. The Conditional Shapley Value Index (CSVI) is proposed based on causal graph theory to filter adverse variables, and the Structural Causal Model (SCM) is then adopted to define the statistical estimand for causal effects. The treatment effects are estimated by Doubly Robust Learning (DRL) methods, which combine doubly robust causal inference with classification and regression machine learning models. Experimental results from 4815 crashes on Highway Interstate 5 in Washington State reveal the heterogeneous treatment effects of crashes at varying distances and durations. The rear-end crashes cause more severe congestion and longer durations than other types of crashes, and the sideswipe crashes have the longest delayed impact. Additionally, the findings show that rear-end crashes affect traffic greater at night, while crash to objects has the most significant influence during peak hours. Statistical hypothesis tests, error metrics based on matched "counterfactual outcomes", and sensitive analyses are employed for assessment, and the results validate the accuracy and effectiveness of our method.
[ { "created": "Mon, 1 Jan 2024 15:03:14 GMT", "version": "v1" } ]
2024-01-02
[ [ "Li", "Shuang", "" ], [ "Pu", "Ziyuan", "" ], [ "Cui", "Zhiyong", "" ], [ "Lee", "Seunghyeon", "" ], [ "Guo", "Xiucheng", "" ], [ "Ngoduy", "Dong", "" ] ]
Highway traffic crashes exert a considerable impact on both transportation systems and the economy. In this context, accurate and dependable emergency responses are crucial for effective traffic management. However, the influence of crashes on traffic status varies across diverse factors and may be biased due to selection bias. Therefore, there arises a necessity to accurately estimate the heterogeneous causal effects of crashes, thereby providing essential insights to facilitate individual-level emergency decision-making. This paper proposes a novel causal machine learning framework to estimate the causal effect of different types of crashes on highway speed. The Neyman-Rubin Causal Model (RCM) is employed to formulate this problem from a causal perspective. The Conditional Shapley Value Index (CSVI) is proposed based on causal graph theory to filter adverse variables, and the Structural Causal Model (SCM) is then adopted to define the statistical estimand for causal effects. The treatment effects are estimated by Doubly Robust Learning (DRL) methods, which combine doubly robust causal inference with classification and regression machine learning models. Experimental results from 4815 crashes on Highway Interstate 5 in Washington State reveal the heterogeneous treatment effects of crashes at varying distances and durations. The rear-end crashes cause more severe congestion and longer durations than other types of crashes, and the sideswipe crashes have the longest delayed impact. Additionally, the findings show that rear-end crashes affect traffic greater at night, while crash to objects has the most significant influence during peak hours. Statistical hypothesis tests, error metrics based on matched "counterfactual outcomes", and sensitive analyses are employed for assessment, and the results validate the accuracy and effectiveness of our method.
2211.13853
Sutanay Choudhury
Hatem Helal, Jesun Firoz, Jenna Bilbrey, Mario Michael Krell, Tom Murray, Ang Li, Sotiris Xantheas, Sutanay Choudhury
Extreme Acceleration of Graph Neural Network-based Prediction Models for Quantum Chemistry
null
null
null
null
cs.LG cs.AR physics.chem-ph
http://creativecommons.org/licenses/by/4.0/
Molecular property calculations are the bedrock of chemical physics. High-fidelity \textit{ab initio} modeling techniques for computing the molecular properties can be prohibitively expensive, and motivate the development of machine-learning models that make the same predictions more efficiently. Training graph neural networks over large molecular databases introduces unique computational challenges such as the need to process millions of small graphs with variable size and support communication patterns that are distinct from learning over large graphs such as social networks. This paper demonstrates a novel hardware-software co-design approach to scale up the training of graph neural networks for molecular property prediction. We introduce an algorithm to coalesce the batches of molecular graphs into fixed size packs to eliminate redundant computation and memory associated with alternative padding techniques and improve throughput via minimizing communication. We demonstrate the effectiveness of our co-design approach by providing an implementation of a well-established molecular property prediction model on the Graphcore Intelligence Processing Units (IPU). We evaluate the training performance on multiple molecular graph databases with varying degrees of graph counts, sizes and sparsity. We demonstrate that such a co-design approach can reduce the training time of such molecular property prediction models from days to less than two hours, opening new possibilities for AI-driven scientific discovery.
[ { "created": "Fri, 25 Nov 2022 01:30:18 GMT", "version": "v1" } ]
2022-11-28
[ [ "Helal", "Hatem", "" ], [ "Firoz", "Jesun", "" ], [ "Bilbrey", "Jenna", "" ], [ "Krell", "Mario Michael", "" ], [ "Murray", "Tom", "" ], [ "Li", "Ang", "" ], [ "Xantheas", "Sotiris", "" ], [ "Choudhury", "Sutanay", "" ] ]
Molecular property calculations are the bedrock of chemical physics. High-fidelity \textit{ab initio} modeling techniques for computing the molecular properties can be prohibitively expensive, and motivate the development of machine-learning models that make the same predictions more efficiently. Training graph neural networks over large molecular databases introduces unique computational challenges such as the need to process millions of small graphs with variable size and support communication patterns that are distinct from learning over large graphs such as social networks. This paper demonstrates a novel hardware-software co-design approach to scale up the training of graph neural networks for molecular property prediction. We introduce an algorithm to coalesce the batches of molecular graphs into fixed size packs to eliminate redundant computation and memory associated with alternative padding techniques and improve throughput via minimizing communication. We demonstrate the effectiveness of our co-design approach by providing an implementation of a well-established molecular property prediction model on the Graphcore Intelligence Processing Units (IPU). We evaluate the training performance on multiple molecular graph databases with varying degrees of graph counts, sizes and sparsity. We demonstrate that such a co-design approach can reduce the training time of such molecular property prediction models from days to less than two hours, opening new possibilities for AI-driven scientific discovery.
2312.07106
Michael Unterkalmsteiner
Eriks Klotins, Michael Unterkalmsteiner, Panagiota Chatzipetrou, Tony Gorschek, Rafael Prikladnicki, Nirnaya Tripathi, Leandro Bento Pompermaier
A Progression Model of Software Engineering Goals, Challenges, and Practices in Start-Ups
null
IEEE Trans. Software Eng. 47(3): 498-521 (2021)
10.1109/TSE.2019.2900213
null
cs.SE
http://creativecommons.org/licenses/by-nc-sa/4.0/
Context: Software start-ups are emerging as suppliers of innovation and software-intensive products. However, traditional software engineering practices are not evaluated in the context, nor adopted to goals and challenges of start-ups. As a result, there is insufficient support for software engineering in the start-up context. Objective: We aim to collect data related to engineering goals, challenges, and practices in start-up companies to ascertain trends and patterns characterizing engineering work in start-ups. Such data allows researchers to understand better how goals and challenges are related to practices. This understanding can then inform future studies aimed at designing solutions addressing those goals and challenges. Besides, these trends and patterns can be useful for practitioners to make more informed decisions in their engineering practice. Method: We use a case survey method to gather first-hand, in-depth experiences from a large sample of software start-ups. We use open coding and cross-case analysis to describe and identify patterns, and corroborate the findings with statistical analysis. Results: We analyze 84 start-up cases and identify 16 goals, 9 challenges, and 16 engineering practices that are common among start-ups. We have mapped these goals, challenges, and practices to start-up life-cycle stages (inception, stabilization, growth, and maturity). Thus, creating the progression model guiding software engineering efforts in start-ups. Conclusions: We conclude that start-ups to a large extent face the same challenges and use the same practices as established companies. However, the primary software engineering challenge in start-ups is to evolve multiple process areas at once, with a little margin for serious errors.
[ { "created": "Tue, 12 Dec 2023 09:36:43 GMT", "version": "v1" } ]
2023-12-13
[ [ "Klotins", "Eriks", "" ], [ "Unterkalmsteiner", "Michael", "" ], [ "Chatzipetrou", "Panagiota", "" ], [ "Gorschek", "Tony", "" ], [ "Prikladnicki", "Rafael", "" ], [ "Tripathi", "Nirnaya", "" ], [ "Pompermaier", "Leandro Bento", "" ] ]
Context: Software start-ups are emerging as suppliers of innovation and software-intensive products. However, traditional software engineering practices are not evaluated in the context, nor adopted to goals and challenges of start-ups. As a result, there is insufficient support for software engineering in the start-up context. Objective: We aim to collect data related to engineering goals, challenges, and practices in start-up companies to ascertain trends and patterns characterizing engineering work in start-ups. Such data allows researchers to understand better how goals and challenges are related to practices. This understanding can then inform future studies aimed at designing solutions addressing those goals and challenges. Besides, these trends and patterns can be useful for practitioners to make more informed decisions in their engineering practice. Method: We use a case survey method to gather first-hand, in-depth experiences from a large sample of software start-ups. We use open coding and cross-case analysis to describe and identify patterns, and corroborate the findings with statistical analysis. Results: We analyze 84 start-up cases and identify 16 goals, 9 challenges, and 16 engineering practices that are common among start-ups. We have mapped these goals, challenges, and practices to start-up life-cycle stages (inception, stabilization, growth, and maturity). Thus, creating the progression model guiding software engineering efforts in start-ups. Conclusions: We conclude that start-ups to a large extent face the same challenges and use the same practices as established companies. However, the primary software engineering challenge in start-ups is to evolve multiple process areas at once, with a little margin for serious errors.
1811.06992
Chris Ying
Chris Ying, Sameer Kumar, Dehao Chen, Tao Wang, Youlong Cheng
Image Classification at Supercomputer Scale
Presented as part of Systems for ML Workshop @ NIPS 2018
null
null
null
cs.LG cs.DC stat.ML
http://creativecommons.org/licenses/by-nc-sa/4.0/
Deep learning is extremely computationally intensive, and hardware vendors have responded by building faster accelerators in large clusters. Training deep learning models at petaFLOPS scale requires overcoming both algorithmic and systems software challenges. In this paper, we discuss three systems-related optimizations: (1) distributed batch normalization to control per-replica batch sizes, (2) input pipeline optimizations to sustain model throughput, and (3) 2-D torus all-reduce to speed up gradient summation. We combine these optimizations to train ResNet-50 on ImageNet to 76.3% accuracy in 2.2 minutes on a 1024-chip TPU v3 Pod with a training throughput of over 1.05 million images/second and no accuracy drop.
[ { "created": "Fri, 16 Nov 2018 19:01:40 GMT", "version": "v1" }, { "created": "Sun, 2 Dec 2018 01:30:42 GMT", "version": "v2" } ]
2018-12-04
[ [ "Ying", "Chris", "" ], [ "Kumar", "Sameer", "" ], [ "Chen", "Dehao", "" ], [ "Wang", "Tao", "" ], [ "Cheng", "Youlong", "" ] ]
Deep learning is extremely computationally intensive, and hardware vendors have responded by building faster accelerators in large clusters. Training deep learning models at petaFLOPS scale requires overcoming both algorithmic and systems software challenges. In this paper, we discuss three systems-related optimizations: (1) distributed batch normalization to control per-replica batch sizes, (2) input pipeline optimizations to sustain model throughput, and (3) 2-D torus all-reduce to speed up gradient summation. We combine these optimizations to train ResNet-50 on ImageNet to 76.3% accuracy in 2.2 minutes on a 1024-chip TPU v3 Pod with a training throughput of over 1.05 million images/second and no accuracy drop.
2303.05919
Yuxin Su
Zhilu Lian, Yangzi Li, Zhixiang Chen, Shiwen Shan, Baoxin Han, Yuxin Su
eBPF-based Working Set Size Estimation in Memory Management
8 pages, 6 figures
null
10.1109/ICSS55994.2022.00036
null
cs.PF cs.AR cs.LG
http://creativecommons.org/licenses/by/4.0/
Working set size estimation (WSS) is of great significance to improve the efficiency of program executing and memory arrangement in modern operating systems. Previous work proposed several methods to estimate WSS, including self-balloning, Zballoning and so on. However, these methods which are based on virtual machine usually cause a large overhead. Thus, using those methods to estimate WSS is impractical. In this paper, we propose a novel framework to efficiently estimate WSS with eBPF (extended Berkeley Packet Filter), a cutting-edge technology which monitors and filters data by being attached to the kernel. With an eBPF program pinned into the kernel, we get the times of page fault and other information of memory allocation. Moreover, we collect WSS via vanilla tool to train a predictive model to complete estimation work with LightGBM, a useful tool which performs well on generating decision trees over continuous value. The experimental results illustrate that our framework can estimate WSS precisely with 98.5\% reduction in overhead compared to traditional methods.
[ { "created": "Tue, 17 Jan 2023 03:12:35 GMT", "version": "v1" } ]
2023-03-13
[ [ "Lian", "Zhilu", "" ], [ "Li", "Yangzi", "" ], [ "Chen", "Zhixiang", "" ], [ "Shan", "Shiwen", "" ], [ "Han", "Baoxin", "" ], [ "Su", "Yuxin", "" ] ]
Working set size estimation (WSS) is of great significance to improve the efficiency of program executing and memory arrangement in modern operating systems. Previous work proposed several methods to estimate WSS, including self-balloning, Zballoning and so on. However, these methods which are based on virtual machine usually cause a large overhead. Thus, using those methods to estimate WSS is impractical. In this paper, we propose a novel framework to efficiently estimate WSS with eBPF (extended Berkeley Packet Filter), a cutting-edge technology which monitors and filters data by being attached to the kernel. With an eBPF program pinned into the kernel, we get the times of page fault and other information of memory allocation. Moreover, we collect WSS via vanilla tool to train a predictive model to complete estimation work with LightGBM, a useful tool which performs well on generating decision trees over continuous value. The experimental results illustrate that our framework can estimate WSS precisely with 98.5\% reduction in overhead compared to traditional methods.
1208.4632
EPTCS
Minas Charalambides (University of Illinois at Urbana-Champaign), Peter Dinges (University of Illinois at Urbana-Champaign), Gul Agha (University of Illinois at Urbana-Champaign)
Parameterized Concurrent Multi-Party Session Types
In Proceedings FOCLASA 2012, arXiv:1208.4327
EPTCS 91, 2012, pp. 16-30
10.4204/EPTCS.91.2
null
cs.PL cs.DC
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
Session types have been proposed as a means of statically verifying implementations of communication protocols. Although prior work has been successful in verifying some classes of protocols, it does not cope well with parameterized, multi-actor scenarios with inherent asynchrony. For example, the sliding window protocol is inexpressible in previously proposed session type systems. This paper describes System-A, a new typing language which overcomes many of the expressiveness limitations of prior work. System-A explicitly supports asynchrony and parallelism, as well as multiple forms of parameterization. We define System-A and show how it can be used for the static verification of a large class of asynchronous communication protocols.
[ { "created": "Wed, 22 Aug 2012 21:58:46 GMT", "version": "v1" } ]
2012-08-24
[ [ "Charalambides", "Minas", "", "University of Illinois at Urbana-Champaign" ], [ "Dinges", "Peter", "", "University of Illinois at Urbana-Champaign" ], [ "Agha", "Gul", "", "University of Illinois at Urbana-Champaign" ] ]
Session types have been proposed as a means of statically verifying implementations of communication protocols. Although prior work has been successful in verifying some classes of protocols, it does not cope well with parameterized, multi-actor scenarios with inherent asynchrony. For example, the sliding window protocol is inexpressible in previously proposed session type systems. This paper describes System-A, a new typing language which overcomes many of the expressiveness limitations of prior work. System-A explicitly supports asynchrony and parallelism, as well as multiple forms of parameterization. We define System-A and show how it can be used for the static verification of a large class of asynchronous communication protocols.
0801.2588
K. Raj Kumar
K. Raj Kumar and Giuseppe Caire
Coding and Decoding for the Dynamic Decode and Forward Relay Protocol
Submitted to the IEEE Transactions on Information Theory
null
null
null
cs.IT math.IT
null
We study the Dynamic Decode and Forward (DDF) protocol for a single half-duplex relay, single-antenna channel with quasi-static fading. The DDF protocol is well-known and has been analyzed in terms of the Diversity-Multiplexing Tradeoff (DMT) in the infinite block length limit. We characterize the finite block length DMT and give new explicit code constructions. The finite block length analysis illuminates a few key aspects that have been neglected in the previous literature: 1) we show that one dominating cause of degradation with respect to the infinite block length regime is the event of decoding error at the relay; 2) we explicitly take into account the fact that the destination does not generally know a priori the relay decision time at which the relay switches from listening to transmit mode. Both the above problems can be tackled by a careful design of the decoding algorithm. In particular, we introduce a decision rejection criterion at the relay based on Forney's decision rule (a variant of the Neyman-Pearson rule), such that the relay triggers transmission only when its decision is reliable. Also, we show that a receiver based on the Generalized Likelihood Ratio Test rule that jointly decodes the relay decision time and the information message achieves the optimal DMT. Our results show that no cyclic redundancy check (CRC) for error detection or additional protocol overhead to communicate the decision time are needed for DDF. Finally, we investigate the use of minimum mean squared error generalized decision feedback equalizer (MMSE-GDFE) lattice decoding at both the relay and the destination, and show that it provides near optimal performance at moderate complexity.
[ { "created": "Wed, 16 Jan 2008 23:05:12 GMT", "version": "v1" } ]
2008-01-18
[ [ "Kumar", "K. Raj", "" ], [ "Caire", "Giuseppe", "" ] ]
We study the Dynamic Decode and Forward (DDF) protocol for a single half-duplex relay, single-antenna channel with quasi-static fading. The DDF protocol is well-known and has been analyzed in terms of the Diversity-Multiplexing Tradeoff (DMT) in the infinite block length limit. We characterize the finite block length DMT and give new explicit code constructions. The finite block length analysis illuminates a few key aspects that have been neglected in the previous literature: 1) we show that one dominating cause of degradation with respect to the infinite block length regime is the event of decoding error at the relay; 2) we explicitly take into account the fact that the destination does not generally know a priori the relay decision time at which the relay switches from listening to transmit mode. Both the above problems can be tackled by a careful design of the decoding algorithm. In particular, we introduce a decision rejection criterion at the relay based on Forney's decision rule (a variant of the Neyman-Pearson rule), such that the relay triggers transmission only when its decision is reliable. Also, we show that a receiver based on the Generalized Likelihood Ratio Test rule that jointly decodes the relay decision time and the information message achieves the optimal DMT. Our results show that no cyclic redundancy check (CRC) for error detection or additional protocol overhead to communicate the decision time are needed for DDF. Finally, we investigate the use of minimum mean squared error generalized decision feedback equalizer (MMSE-GDFE) lattice decoding at both the relay and the destination, and show that it provides near optimal performance at moderate complexity.
2204.04717
Kheeran K. Naidu
Cezar-Mihail Alexandru, Pavel Dvo\v{r}\'ak, Christian Konrad, Kheeran K. Naidu
Improved Weighted Matching in the Sliding Window Model
null
null
null
null
cs.DS
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
We consider the Maximum-weight Matching (MWM) problem in the streaming sliding window model of computation. In this model, the input consists of a sequence of weighted edges on a given vertex set $V$ of size $n$. The objective is to maintain an approximation of a maximum-weight matching in the graph spanned by the $L$ most recent edges, for some integer $L$, using as little space as possible. Prior to our work, the state-of-the-art results were a $(3.5+\varepsilon)$-approximation algorithm for MWM by Biabani et al. [ISAAC'21] and a $(3+\varepsilon)$-approximation for (unweighted) Maximum Matching (MM) by Crouch et al. [ESA'13]. Both algorithms use space $\tilde{O}(n)$. We give the following results: 1. We give a $(2+\varepsilon)$-approximation algorithm for MWM with space $\tilde{O}(\sqrt{nL})$. Under the reasonable assumption that the graphs spanned by the edges in each sliding window are simple, our algorithm uses space $\tilde{O}(n \sqrt{n})$. 2. In the $\tilde{O}(n)$ space regime, we give a $(3+\varepsilon)$-approximation algorithm for MWM, thereby closing the gap between the best-known approximation ratio for MWM and MM. Similar to Biabani et al.'s MWM algorithm, both our algorithms execute multiple instances of the $(2+\varepsilon)$-approximation $\tilde{O}(n)$-space streaming algorithm for MWM by Paz and Schwartzman [SODA'17] on different portions of the stream. Our improvements are obtained by selecting these substreams differently. Furthermore, our $(2+\varepsilon)$-approximation algorithm runs the Paz-Schwartzman algorithm in reverse direction over some parts of the stream, and in forward direction over other parts, which allows for an improved approximation guarantee at the cost of increased space requirements.
[ { "created": "Sun, 10 Apr 2022 16:26:11 GMT", "version": "v1" }, { "created": "Tue, 10 Jan 2023 15:22:03 GMT", "version": "v2" } ]
2023-01-11
[ [ "Alexandru", "Cezar-Mihail", "" ], [ "Dvořák", "Pavel", "" ], [ "Konrad", "Christian", "" ], [ "Naidu", "Kheeran K.", "" ] ]
We consider the Maximum-weight Matching (MWM) problem in the streaming sliding window model of computation. In this model, the input consists of a sequence of weighted edges on a given vertex set $V$ of size $n$. The objective is to maintain an approximation of a maximum-weight matching in the graph spanned by the $L$ most recent edges, for some integer $L$, using as little space as possible. Prior to our work, the state-of-the-art results were a $(3.5+\varepsilon)$-approximation algorithm for MWM by Biabani et al. [ISAAC'21] and a $(3+\varepsilon)$-approximation for (unweighted) Maximum Matching (MM) by Crouch et al. [ESA'13]. Both algorithms use space $\tilde{O}(n)$. We give the following results: 1. We give a $(2+\varepsilon)$-approximation algorithm for MWM with space $\tilde{O}(\sqrt{nL})$. Under the reasonable assumption that the graphs spanned by the edges in each sliding window are simple, our algorithm uses space $\tilde{O}(n \sqrt{n})$. 2. In the $\tilde{O}(n)$ space regime, we give a $(3+\varepsilon)$-approximation algorithm for MWM, thereby closing the gap between the best-known approximation ratio for MWM and MM. Similar to Biabani et al.'s MWM algorithm, both our algorithms execute multiple instances of the $(2+\varepsilon)$-approximation $\tilde{O}(n)$-space streaming algorithm for MWM by Paz and Schwartzman [SODA'17] on different portions of the stream. Our improvements are obtained by selecting these substreams differently. Furthermore, our $(2+\varepsilon)$-approximation algorithm runs the Paz-Schwartzman algorithm in reverse direction over some parts of the stream, and in forward direction over other parts, which allows for an improved approximation guarantee at the cost of increased space requirements.
2405.02047
Martin Kumm
Andreas B\"ottcher, Martin Kumm
Small Logic-based Multipliers with Incomplete Sub-Multipliers for FPGAs
Preprint, to appear at ARITH 2024 (http://arith24.arithsymposium.org) and IEEEXplore
null
null
null
cs.AR
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
There is a recent trend in artificial intelligence (AI) inference towards lower precision data formats down to 8 bits and less. As multiplication is the most complex operation in typical inference tasks, there is a large demand for efficient small multipliers. The large DSP blocks have limitations implementing many small multipliers efficiently. Hence, this work proposes a solution for better logic-based multipliers that is especially beneficial for small multipliers. Our work is based on the multiplier tiling method in which a multiplier is designed out of several sub-multiplier tiles. The key observation we made is that these sub-multipliers do not necessarily have to perform a complete (rectangular) NxK multiplication and more efficient sub-multipliers are possible that are incomplete (non-rectangular). This proposal first seeks to identify efficient incomplete irregular sub-multipliers and then demonstrates improvements over state-of-the-art designs. It is shown that optimal solutions can be found using integer linear programming (ILP), which are evaluated in FPGA synthesis experiments.
[ { "created": "Fri, 3 May 2024 12:29:07 GMT", "version": "v1" } ]
2024-05-06
[ [ "Böttcher", "Andreas", "" ], [ "Kumm", "Martin", "" ] ]
There is a recent trend in artificial intelligence (AI) inference towards lower precision data formats down to 8 bits and less. As multiplication is the most complex operation in typical inference tasks, there is a large demand for efficient small multipliers. The large DSP blocks have limitations implementing many small multipliers efficiently. Hence, this work proposes a solution for better logic-based multipliers that is especially beneficial for small multipliers. Our work is based on the multiplier tiling method in which a multiplier is designed out of several sub-multiplier tiles. The key observation we made is that these sub-multipliers do not necessarily have to perform a complete (rectangular) NxK multiplication and more efficient sub-multipliers are possible that are incomplete (non-rectangular). This proposal first seeks to identify efficient incomplete irregular sub-multipliers and then demonstrates improvements over state-of-the-art designs. It is shown that optimal solutions can be found using integer linear programming (ILP), which are evaluated in FPGA synthesis experiments.
1905.10464
Mamoru Komachi
Tosho Hirasawa and Mamoru Komachi
Debiasing Word Embeddings Improves Multimodal Machine Translation
11 pages; MT Summit 2019 (camera ready)
null
null
null
cs.CL
http://creativecommons.org/licenses/by/4.0/
In recent years, pretrained word embeddings have proved useful for multimodal neural machine translation (NMT) models to address the shortage of available datasets. However, the integration of pretrained word embeddings has not yet been explored extensively. Further, pretrained word embeddings in high dimensional spaces have been reported to suffer from the hubness problem. Although some debiasing techniques have been proposed to address this problem for other natural language processing tasks, they have seldom been studied for multimodal NMT models. In this study, we examine various kinds of word embeddings and introduce two debiasing techniques for three multimodal NMT models and two language pairs -- English-German translation and English-French translation. With our optimal settings, the overall performance of multimodal models was improved by up to +1.93 BLEU and +2.02 METEOR for English-German translation and +1.73 BLEU and +0.95 METEOR for English-French translation.
[ { "created": "Fri, 24 May 2019 22:11:57 GMT", "version": "v1" }, { "created": "Tue, 28 May 2019 22:46:58 GMT", "version": "v2" }, { "created": "Sat, 22 Jun 2019 07:50:03 GMT", "version": "v3" } ]
2019-06-25
[ [ "Hirasawa", "Tosho", "" ], [ "Komachi", "Mamoru", "" ] ]
In recent years, pretrained word embeddings have proved useful for multimodal neural machine translation (NMT) models to address the shortage of available datasets. However, the integration of pretrained word embeddings has not yet been explored extensively. Further, pretrained word embeddings in high dimensional spaces have been reported to suffer from the hubness problem. Although some debiasing techniques have been proposed to address this problem for other natural language processing tasks, they have seldom been studied for multimodal NMT models. In this study, we examine various kinds of word embeddings and introduce two debiasing techniques for three multimodal NMT models and two language pairs -- English-German translation and English-French translation. With our optimal settings, the overall performance of multimodal models was improved by up to +1.93 BLEU and +2.02 METEOR for English-German translation and +1.73 BLEU and +0.95 METEOR for English-French translation.
2311.14762
Benjamin Kiefer
Benjamin Kiefer, Lojze \v{Z}ust, Matej Kristan, Janez Per\v{s}, Matija Ter\v{s}ek, Arnold Wiliem, Martin Messmer, Cheng-Yen Yang, Hsiang-Wei Huang, Zhongyu Jiang, Heng-Cheng Kuo, Jie Mei, Jenq-Neng Hwang, Daniel Stadler, Lars Sommer, Kaer Huang, Aiguo Zheng, Weitu Chong, Kanokphan Lertniphonphan, Jun Xie, Feng Chen, Jian Li, Zhepeng Wang, Luca Zedda, Andrea Loddo, Cecilia Di Ruberto, Tuan-Anh Vu, Hai Nguyen-Truong, Tan-Sang Ha, Quan-Dung Pham, Sai-Kit Yeung, Yuan Feng, Nguyen Thanh Thien, Lixin Tian, Sheng-Yao Kuan, Yuan-Hao Ho, Angel Bueno Rodriguez, Borja Carrillo-Perez, Alexander Klein, Antje Alex, Yannik Steiniger, Felix Sattler, Edgardo Solano-Carrillo, Matej Fabijani\'c, Magdalena \v{S}umunec, Nadir Kapetanovi\'c, Andreas Michel, Wolfgang Gross, Martin Weinmann
The 2nd Workshop on Maritime Computer Vision (MaCVi) 2024
Part of 2nd Workshop on Maritime Computer Vision (MaCVi) 2024 IEEE Xplore submission as part of WACV 2024
null
null
null
cs.CV cs.AI
http://creativecommons.org/licenses/by/4.0/
The 2nd Workshop on Maritime Computer Vision (MaCVi) 2024 addresses maritime computer vision for Unmanned Aerial Vehicles (UAV) and Unmanned Surface Vehicles (USV). Three challenges categories are considered: (i) UAV-based Maritime Object Tracking with Re-identification, (ii) USV-based Maritime Obstacle Segmentation and Detection, (iii) USV-based Maritime Boat Tracking. The USV-based Maritime Obstacle Segmentation and Detection features three sub-challenges, including a new embedded challenge addressing efficicent inference on real-world embedded devices. This report offers a comprehensive overview of the findings from the challenges. We provide both statistical and qualitative analyses, evaluating trends from over 195 submissions. All datasets, evaluation code, and the leaderboard are available to the public at https://macvi.org/workshop/macvi24.
[ { "created": "Thu, 23 Nov 2023 21:01:14 GMT", "version": "v1" } ]
2023-11-28
[ [ "Kiefer", "Benjamin", "" ], [ "Žust", "Lojze", "" ], [ "Kristan", "Matej", "" ], [ "Perš", "Janez", "" ], [ "Teršek", "Matija", "" ], [ "Wiliem", "Arnold", "" ], [ "Messmer", "Martin", "" ], [ "Yang", "Cheng-Yen", "" ], [ "Huang", "Hsiang-Wei", "" ], [ "Jiang", "Zhongyu", "" ], [ "Kuo", "Heng-Cheng", "" ], [ "Mei", "Jie", "" ], [ "Hwang", "Jenq-Neng", "" ], [ "Stadler", "Daniel", "" ], [ "Sommer", "Lars", "" ], [ "Huang", "Kaer", "" ], [ "Zheng", "Aiguo", "" ], [ "Chong", "Weitu", "" ], [ "Lertniphonphan", "Kanokphan", "" ], [ "Xie", "Jun", "" ], [ "Chen", "Feng", "" ], [ "Li", "Jian", "" ], [ "Wang", "Zhepeng", "" ], [ "Zedda", "Luca", "" ], [ "Loddo", "Andrea", "" ], [ "Di Ruberto", "Cecilia", "" ], [ "Vu", "Tuan-Anh", "" ], [ "Nguyen-Truong", "Hai", "" ], [ "Ha", "Tan-Sang", "" ], [ "Pham", "Quan-Dung", "" ], [ "Yeung", "Sai-Kit", "" ], [ "Feng", "Yuan", "" ], [ "Thien", "Nguyen Thanh", "" ], [ "Tian", "Lixin", "" ], [ "Kuan", "Sheng-Yao", "" ], [ "Ho", "Yuan-Hao", "" ], [ "Rodriguez", "Angel Bueno", "" ], [ "Carrillo-Perez", "Borja", "" ], [ "Klein", "Alexander", "" ], [ "Alex", "Antje", "" ], [ "Steiniger", "Yannik", "" ], [ "Sattler", "Felix", "" ], [ "Solano-Carrillo", "Edgardo", "" ], [ "Fabijanić", "Matej", "" ], [ "Šumunec", "Magdalena", "" ], [ "Kapetanović", "Nadir", "" ], [ "Michel", "Andreas", "" ], [ "Gross", "Wolfgang", "" ], [ "Weinmann", "Martin", "" ] ]
The 2nd Workshop on Maritime Computer Vision (MaCVi) 2024 addresses maritime computer vision for Unmanned Aerial Vehicles (UAV) and Unmanned Surface Vehicles (USV). Three challenges categories are considered: (i) UAV-based Maritime Object Tracking with Re-identification, (ii) USV-based Maritime Obstacle Segmentation and Detection, (iii) USV-based Maritime Boat Tracking. The USV-based Maritime Obstacle Segmentation and Detection features three sub-challenges, including a new embedded challenge addressing efficicent inference on real-world embedded devices. This report offers a comprehensive overview of the findings from the challenges. We provide both statistical and qualitative analyses, evaluating trends from over 195 submissions. All datasets, evaluation code, and the leaderboard are available to the public at https://macvi.org/workshop/macvi24.
2107.14388
Yongxiang Gu
Yongxiang Gu, Qianlei Wang, Xiaolin Qin
Real-time Streaming Perception System for Autonomous Driving
6 pages,6 figures
null
null
null
cs.CV
http://creativecommons.org/licenses/by/4.0/
Nowadays, plenty of deep learning technologies are being applied to all aspects of autonomous driving with promising results. Among them, object detection is the key to improve the ability of an autonomous agent to perceive its environment so that it can (re)act. However, previous vision-based object detectors cannot achieve satisfactory performance under real-time driving scenarios. To remedy this, we present the real-time steaming perception system in this paper, which is also the 2nd Place solution of Streaming Perception Challenge (Workshop on Autonomous Driving at CVPR 2021) for the detection-only track. Unlike traditional object detection challenges, which focus mainly on the absolute performance, streaming perception task requires achieving a balance of accuracy and latency, which is crucial for real-time autonomous driving. We adopt YOLOv5 as our basic framework, data augmentation, Bag-of-Freebies, and Transformer are adopted to improve streaming object detection performance with negligible extra inference cost. On the Argoverse-HD test set, our method achieves 33.2 streaming AP (34.6 streaming AP verified by the organizer) under the required hardware. Its performance significantly surpasses the fixed baseline of 13.6 (host team), demonstrating the potentiality of application.
[ { "created": "Fri, 30 Jul 2021 01:32:44 GMT", "version": "v1" } ]
2021-08-02
[ [ "Gu", "Yongxiang", "" ], [ "Wang", "Qianlei", "" ], [ "Qin", "Xiaolin", "" ] ]
Nowadays, plenty of deep learning technologies are being applied to all aspects of autonomous driving with promising results. Among them, object detection is the key to improve the ability of an autonomous agent to perceive its environment so that it can (re)act. However, previous vision-based object detectors cannot achieve satisfactory performance under real-time driving scenarios. To remedy this, we present the real-time steaming perception system in this paper, which is also the 2nd Place solution of Streaming Perception Challenge (Workshop on Autonomous Driving at CVPR 2021) for the detection-only track. Unlike traditional object detection challenges, which focus mainly on the absolute performance, streaming perception task requires achieving a balance of accuracy and latency, which is crucial for real-time autonomous driving. We adopt YOLOv5 as our basic framework, data augmentation, Bag-of-Freebies, and Transformer are adopted to improve streaming object detection performance with negligible extra inference cost. On the Argoverse-HD test set, our method achieves 33.2 streaming AP (34.6 streaming AP verified by the organizer) under the required hardware. Its performance significantly surpasses the fixed baseline of 13.6 (host team), demonstrating the potentiality of application.
1701.02009
Alexander Zhdanov
Alexander Zhdanov
IRA codes derived from Gruenbaum graph
null
null
null
null
cs.IT math.IT
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
In this paper, we consider coding of short data frames (192 bits) by IRA codes. A new interleaver for the IRA codes based on a Gruenbaum graph is proposed. The difference of the proposed algorithm from known methods consists in the following: permutation is performed by using a match smaller interleaver which is derived from the Gruenbaum graph by finding in this graph a Hamiltonian path, enumerating the passed vertices in ascending order and passing them again in the permuted order through the edges which are not included in the Hamiltonian path. For the IRA code the obtained interleaver provides 0.7-0.8 db gain over a convolutional code decoded by Viterbi algorithm.
[ { "created": "Sun, 8 Jan 2017 19:59:35 GMT", "version": "v1" } ]
2017-01-10
[ [ "Zhdanov", "Alexander", "" ] ]
In this paper, we consider coding of short data frames (192 bits) by IRA codes. A new interleaver for the IRA codes based on a Gruenbaum graph is proposed. The difference of the proposed algorithm from known methods consists in the following: permutation is performed by using a match smaller interleaver which is derived from the Gruenbaum graph by finding in this graph a Hamiltonian path, enumerating the passed vertices in ascending order and passing them again in the permuted order through the edges which are not included in the Hamiltonian path. For the IRA code the obtained interleaver provides 0.7-0.8 db gain over a convolutional code decoded by Viterbi algorithm.
2312.06875
Ryan Beckett
Siva Kesava Reddy Kakarla, Ryan Beckett
Oracle-based Protocol Testing with Eywa
null
null
null
null
cs.NI
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
We present oracle-based testing a new technique for automatic black-box testing of network protocol implementations. Oracle-based testing leverages recent advances in LLMs to build rich models of intended protocol behavior from knowledge embedded in RFCs, blogs, forums, and other natural language sources. From these models it systematically derives exhaustive test cases using symbolic program execution. We realize oracle-based testing through Eywa, a novel protocol testing framework implemented in Python. To demonstrate Eywa's effectiveness, we show its use through an extensive case study of the DNS protocol. Despite requiring minimal effort, applying Eywa to the DNS resulting in the discovery of 26 unique bugs across ten widely used DNS implementations, including 11 new bugs that were previously undiscovered despite elaborate prior testing with manually crafted models.
[ { "created": "Mon, 11 Dec 2023 22:51:15 GMT", "version": "v1" } ]
2023-12-13
[ [ "Kakarla", "Siva Kesava Reddy", "" ], [ "Beckett", "Ryan", "" ] ]
We present oracle-based testing a new technique for automatic black-box testing of network protocol implementations. Oracle-based testing leverages recent advances in LLMs to build rich models of intended protocol behavior from knowledge embedded in RFCs, blogs, forums, and other natural language sources. From these models it systematically derives exhaustive test cases using symbolic program execution. We realize oracle-based testing through Eywa, a novel protocol testing framework implemented in Python. To demonstrate Eywa's effectiveness, we show its use through an extensive case study of the DNS protocol. Despite requiring minimal effort, applying Eywa to the DNS resulting in the discovery of 26 unique bugs across ten widely used DNS implementations, including 11 new bugs that were previously undiscovered despite elaborate prior testing with manually crafted models.
cs/0410004
Andras Lorincz
I. Szita and A. Lorincz
Applying Policy Iteration for Training Recurrent Neural Networks
Supplementary material. 17 papes, 1 figure
null
null
null
cs.AI cs.LG cs.NE
null
Recurrent neural networks are often used for learning time-series data. Based on a few assumptions we model this learning task as a minimization problem of a nonlinear least-squares cost function. The special structure of the cost function allows us to build a connection to reinforcement learning. We exploit this connection and derive a convergent, policy iteration-based algorithm. Furthermore, we argue that RNN training can be fit naturally into the reinforcement learning framework.
[ { "created": "Sat, 2 Oct 2004 07:19:49 GMT", "version": "v1" } ]
2007-05-23
[ [ "Szita", "I.", "" ], [ "Lorincz", "A.", "" ] ]
Recurrent neural networks are often used for learning time-series data. Based on a few assumptions we model this learning task as a minimization problem of a nonlinear least-squares cost function. The special structure of the cost function allows us to build a connection to reinforcement learning. We exploit this connection and derive a convergent, policy iteration-based algorithm. Furthermore, we argue that RNN training can be fit naturally into the reinforcement learning framework.
2405.16487
Tyler Han
Tyler Han, Sidharth Talia, Rohan Panicker, Preet Shah, Neel Jawale, Byron Boots
Dynamics Models in the Aggressive Off-Road Driving Regime
Accepted to ICRA 2024 Workshop on Resilient Off-road Autonomy
null
null
null
cs.RO
http://creativecommons.org/licenses/by/4.0/
Current developments in autonomous off-road driving are steadily increasing performance through higher speeds and more challenging, unstructured environments. However, this operating regime subjects the vehicle to larger inertial effects, where consideration of higher-order states is necessary to avoid failures such as rollovers or excessive impact forces. Aggressive driving through Model Predictive Control (MPC) in these conditions requires dynamics models that accurately predict safety-critical information. This work aims to empirically quantify this aggressive operating regime and its effects on the performance of current models. We evaluate three dynamics models of varying complexity on two distinct off-road driving datasets: one simulated and the other real-world. By conditioning trajectory data on higher-order states, we show that model accuracy degrades with aggressiveness and simpler models degrade faster. These models are also validated across datasets, where accuracies over safety-critical states are reported and provide benchmarks for future work.
[ { "created": "Sun, 26 May 2024 08:52:16 GMT", "version": "v1" } ]
2024-05-28
[ [ "Han", "Tyler", "" ], [ "Talia", "Sidharth", "" ], [ "Panicker", "Rohan", "" ], [ "Shah", "Preet", "" ], [ "Jawale", "Neel", "" ], [ "Boots", "Byron", "" ] ]
Current developments in autonomous off-road driving are steadily increasing performance through higher speeds and more challenging, unstructured environments. However, this operating regime subjects the vehicle to larger inertial effects, where consideration of higher-order states is necessary to avoid failures such as rollovers or excessive impact forces. Aggressive driving through Model Predictive Control (MPC) in these conditions requires dynamics models that accurately predict safety-critical information. This work aims to empirically quantify this aggressive operating regime and its effects on the performance of current models. We evaluate three dynamics models of varying complexity on two distinct off-road driving datasets: one simulated and the other real-world. By conditioning trajectory data on higher-order states, we show that model accuracy degrades with aggressiveness and simpler models degrade faster. These models are also validated across datasets, where accuracies over safety-critical states are reported and provide benchmarks for future work.
2008.04109
Abdul Mueed Hafiz Dr.
Abdul Mueed Hafiz and Ghulam Mohiuddin Bhat
Deep Q-Network Based Multi-agent Reinforcement Learning with Binary Action Agents
null
null
null
null
cs.LG cs.AI cs.MA cs.SY eess.SY stat.ML
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
Deep Q-Network (DQN) based multi-agent systems (MAS) for reinforcement learning (RL) use various schemes where in the agents have to learn and communicate. The learning is however specific to each agent and communication may be satisfactorily designed for the agents. As more complex Deep QNetworks come to the fore, the overall complexity of the multi-agent system increases leading to issues like difficulty in training, need for higher resources and more training time, difficulty in fine-tuning, etc. To address these issues we propose a simple but efficient DQN based MAS for RL which uses shared state and rewards, but agent-specific actions, for updation of the experience replay pool of the DQNs, where each agent is a DQN. The benefits of the approach are overall simplicity, faster convergence and better performance as compared to conventional DQN based approaches. It should be noted that the method can be extended to any DQN. As such we use simple DQN and DDQN (Double Q-learning) respectively on three separate tasks i.e. Cartpole-v1 (OpenAI Gym environment) , LunarLander-v2 (OpenAI Gym environment) and Maze Traversal (customized environment). The proposed approach outperforms the baseline on these tasks by decent margins respectively.
[ { "created": "Thu, 6 Aug 2020 15:16:05 GMT", "version": "v1" } ]
2020-08-11
[ [ "Hafiz", "Abdul Mueed", "" ], [ "Bhat", "Ghulam Mohiuddin", "" ] ]
Deep Q-Network (DQN) based multi-agent systems (MAS) for reinforcement learning (RL) use various schemes where in the agents have to learn and communicate. The learning is however specific to each agent and communication may be satisfactorily designed for the agents. As more complex Deep QNetworks come to the fore, the overall complexity of the multi-agent system increases leading to issues like difficulty in training, need for higher resources and more training time, difficulty in fine-tuning, etc. To address these issues we propose a simple but efficient DQN based MAS for RL which uses shared state and rewards, but agent-specific actions, for updation of the experience replay pool of the DQNs, where each agent is a DQN. The benefits of the approach are overall simplicity, faster convergence and better performance as compared to conventional DQN based approaches. It should be noted that the method can be extended to any DQN. As such we use simple DQN and DDQN (Double Q-learning) respectively on three separate tasks i.e. Cartpole-v1 (OpenAI Gym environment) , LunarLander-v2 (OpenAI Gym environment) and Maze Traversal (customized environment). The proposed approach outperforms the baseline on these tasks by decent margins respectively.
2202.13057
Vsevolod Nikulin
Vsevolod Nikulin and Jun Tani
Initialization of Latent Space Coordinates via Random Linear Projections for Learning Robotic Sensory-Motor Sequences
18 pages, 9 figures
null
null
null
cs.LG cs.RO
http://creativecommons.org/licenses/by/4.0/
Robot kinematics data, despite being a high dimensional process, is highly correlated, especially when considering motions grouped in certain primitives. These almost linear correlations within primitives allow us to interpret the motions as points drawn close to a union of low-dimensional linear subspaces in the space of all motions. Motivated by results of embedding theory, in particular, generalizations of Whitney embedding theorem, we show that random linear projection of motor sequences into low dimensional space loses very little information about structure of kinematics data. Projected points are very good initial guess for values of latent variables in generative model for robot sensory-motor behaviour primitives. We conducted series of experiments where we trained a recurrent neural network to generate sensory-motor sequences for robotic manipulator with 9 degrees of freedom. Experimental results demonstrate substantial improvement in generalisation abilities for unobserved samples in the case of initialization of latent variables with random linear projection of motor data over initialization with zero or random values. Moreover, latent space is well-structured wherein samples belonging to different primitives are well separated from the onset of training process.
[ { "created": "Sat, 26 Feb 2022 04:32:16 GMT", "version": "v1" } ]
2022-03-01
[ [ "Nikulin", "Vsevolod", "" ], [ "Tani", "Jun", "" ] ]
Robot kinematics data, despite being a high dimensional process, is highly correlated, especially when considering motions grouped in certain primitives. These almost linear correlations within primitives allow us to interpret the motions as points drawn close to a union of low-dimensional linear subspaces in the space of all motions. Motivated by results of embedding theory, in particular, generalizations of Whitney embedding theorem, we show that random linear projection of motor sequences into low dimensional space loses very little information about structure of kinematics data. Projected points are very good initial guess for values of latent variables in generative model for robot sensory-motor behaviour primitives. We conducted series of experiments where we trained a recurrent neural network to generate sensory-motor sequences for robotic manipulator with 9 degrees of freedom. Experimental results demonstrate substantial improvement in generalisation abilities for unobserved samples in the case of initialization of latent variables with random linear projection of motor data over initialization with zero or random values. Moreover, latent space is well-structured wherein samples belonging to different primitives are well separated from the onset of training process.
2405.15092
Evelyn Yee
Evelyn Yee and Alice Li and Chenyu Tang and Yeon Ho Jung and Ramamohan Paturi and Leon Bergen
Dissociation of Faithful and Unfaithful Reasoning in LLMs
code published at https://github.com/CoTErrorRecovery/CoTErrorRecovery
null
null
null
cs.AI cs.CL
http://creativecommons.org/licenses/by/4.0/
Large language models (LLMs) improve their performance in downstream tasks when they generate Chain of Thought reasoning text before producing an answer. Our research investigates how LLMs recover from errors in Chain of Thought, reaching the correct final answer despite mistakes in the reasoning text. Through analysis of these error recovery behaviors, we find evidence for unfaithfulness in Chain of Thought, but we also identify many clear examples of faithful error recovery behaviors. We identify factors that shift LLM recovery behavior: LLMs recover more frequently from obvious errors and in contexts that provide more evidence for the correct answer. However, unfaithful recoveries show the opposite behavior, occurring more frequently for more difficult error positions. Our results indicate that there are distinct mechanisms driving faithful and unfaithful error recoveries. Our results challenge the view that LLM reasoning is a uniform, coherent process.
[ { "created": "Thu, 23 May 2024 22:38:58 GMT", "version": "v1" } ]
2024-05-27
[ [ "Yee", "Evelyn", "" ], [ "Li", "Alice", "" ], [ "Tang", "Chenyu", "" ], [ "Jung", "Yeon Ho", "" ], [ "Paturi", "Ramamohan", "" ], [ "Bergen", "Leon", "" ] ]
Large language models (LLMs) improve their performance in downstream tasks when they generate Chain of Thought reasoning text before producing an answer. Our research investigates how LLMs recover from errors in Chain of Thought, reaching the correct final answer despite mistakes in the reasoning text. Through analysis of these error recovery behaviors, we find evidence for unfaithfulness in Chain of Thought, but we also identify many clear examples of faithful error recovery behaviors. We identify factors that shift LLM recovery behavior: LLMs recover more frequently from obvious errors and in contexts that provide more evidence for the correct answer. However, unfaithful recoveries show the opposite behavior, occurring more frequently for more difficult error positions. Our results indicate that there are distinct mechanisms driving faithful and unfaithful error recoveries. Our results challenge the view that LLM reasoning is a uniform, coherent process.
1706.05893
Simon Wacker
Simon Wacker
Signal Machine And Cellular Automaton Time-Optimal Quasi-Solutions Of The Firing Squad/Mob Synchronisation Problem On Connected Graphs
null
null
null
null
cs.FL cs.DC cs.DS
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
We construct a time-optimal quasi-solution of the firing mob synchronisation problem over finite, connected, and undirected multigraphs whose maximum degrees are uniformly bounded by a constant. It is only a quasi-solution because its number of states depends on the graph or, from another perspective, does not depend on the graph but is countably infinite. To construct this quasi-solution we introduce signal machines over continuum representations of such multigraphs and construct a signal machine whose discretisation is a cellular automaton that quasi-solves the problem. This automaton uses a time-optimal solution of the firing squad synchronisation problem in dimension one with one general at one end to synchronise edges, and freezes and thaws the synchronisation of edges in such a way that all edges synchronise at the same time.
[ { "created": "Mon, 19 Jun 2017 11:47:45 GMT", "version": "v1" } ]
2017-06-20
[ [ "Wacker", "Simon", "" ] ]
We construct a time-optimal quasi-solution of the firing mob synchronisation problem over finite, connected, and undirected multigraphs whose maximum degrees are uniformly bounded by a constant. It is only a quasi-solution because its number of states depends on the graph or, from another perspective, does not depend on the graph but is countably infinite. To construct this quasi-solution we introduce signal machines over continuum representations of such multigraphs and construct a signal machine whose discretisation is a cellular automaton that quasi-solves the problem. This automaton uses a time-optimal solution of the firing squad synchronisation problem in dimension one with one general at one end to synchronise edges, and freezes and thaws the synchronisation of edges in such a way that all edges synchronise at the same time.
2403.00553
Chantal Shaib
Chantal Shaib, Joe Barrow, Jiuding Sun, Alexa F. Siu, Byron C. Wallace, Ani Nenkova
Standardizing the Measurement of Text Diversity: A Tool and a Comparative Analysis of Scores
Preprint
null
null
null
cs.CL
http://creativecommons.org/licenses/by/4.0/
The diversity across outputs generated by large language models shapes the perception of their quality and utility. Prompt leaks, templated answer structure, and canned responses across different interactions are readily noticed by people, but there is no standard score to measure this aspect of model behavior. In this work we empirically investigate diversity scores on English texts. We find that computationally efficient compression algorithms capture information similar to what is measured by slow to compute $n$-gram overlap homogeneity scores. Further, a combination of measures -- compression ratios, self-repetition of long $n$-grams and Self-BLEU and BERTScore -- are sufficient to report, as they have low mutual correlation with each other. The applicability of scores extends beyond analysis of generative models; for example, we highlight applications on instruction-tuning datasets and human-produced texts. We release a diversity score package to facilitate research and invite consistency across reports.
[ { "created": "Fri, 1 Mar 2024 14:23:12 GMT", "version": "v1" } ]
2024-03-04
[ [ "Shaib", "Chantal", "" ], [ "Barrow", "Joe", "" ], [ "Sun", "Jiuding", "" ], [ "Siu", "Alexa F.", "" ], [ "Wallace", "Byron C.", "" ], [ "Nenkova", "Ani", "" ] ]
The diversity across outputs generated by large language models shapes the perception of their quality and utility. Prompt leaks, templated answer structure, and canned responses across different interactions are readily noticed by people, but there is no standard score to measure this aspect of model behavior. In this work we empirically investigate diversity scores on English texts. We find that computationally efficient compression algorithms capture information similar to what is measured by slow to compute $n$-gram overlap homogeneity scores. Further, a combination of measures -- compression ratios, self-repetition of long $n$-grams and Self-BLEU and BERTScore -- are sufficient to report, as they have low mutual correlation with each other. The applicability of scores extends beyond analysis of generative models; for example, we highlight applications on instruction-tuning datasets and human-produced texts. We release a diversity score package to facilitate research and invite consistency across reports.
2401.02861
Marta Gomez-Barrero
Marta Gomez-Barrero, Javier Galbally
Reversing the Irreversible: A Survey on Inverse Biometrics
18 pages, journal, survey
Elsevier Computers & Security, Volume 90, March 2020, 101700
10.1016/j.cose.2019.101700
null
cs.CV
http://creativecommons.org/licenses/by/4.0/
With the widespread use of biometric recognition, several issues related to the privacy and security provided by this technology have been recently raised and analysed. As a result, the early common belief among the biometrics community of templates irreversibility has been proven wrong. It is now an accepted fact that it is possible to reconstruct from an unprotected template a synthetic sample that matches the bona fide one. This reverse engineering process, commonly referred to as \textit{inverse biometrics}, constitutes a severe threat for biometric systems from two different angles: on the one hand, sensitive personal data (i.e., biometric data) can be derived from compromised unprotected templates; on the other hand, other powerful attacks can be launched building upon these reconstructed samples. Given its important implications, biometric stakeholders have produced over the last fifteen years numerous works analysing the different aspects related to inverse biometrics: development of reconstruction algorithms for different characteristics; proposal of methodologies to assess the vulnerabilities of biometric systems to the aforementioned algorithms; development of countermeasures to reduce the possible effects of attacks. The present article is an effort to condense all this information in one comprehensive review of: the problem itself, the evaluation of the problem, and the mitigation of the problem. The present article is an effort to condense all this information in one comprehensive review of: the problem itself, the evaluation of the problem, and the mitigation of the problem.
[ { "created": "Fri, 5 Jan 2024 15:32:40 GMT", "version": "v1" } ]
2024-01-08
[ [ "Gomez-Barrero", "Marta", "" ], [ "Galbally", "Javier", "" ] ]
With the widespread use of biometric recognition, several issues related to the privacy and security provided by this technology have been recently raised and analysed. As a result, the early common belief among the biometrics community of templates irreversibility has been proven wrong. It is now an accepted fact that it is possible to reconstruct from an unprotected template a synthetic sample that matches the bona fide one. This reverse engineering process, commonly referred to as \textit{inverse biometrics}, constitutes a severe threat for biometric systems from two different angles: on the one hand, sensitive personal data (i.e., biometric data) can be derived from compromised unprotected templates; on the other hand, other powerful attacks can be launched building upon these reconstructed samples. Given its important implications, biometric stakeholders have produced over the last fifteen years numerous works analysing the different aspects related to inverse biometrics: development of reconstruction algorithms for different characteristics; proposal of methodologies to assess the vulnerabilities of biometric systems to the aforementioned algorithms; development of countermeasures to reduce the possible effects of attacks. The present article is an effort to condense all this information in one comprehensive review of: the problem itself, the evaluation of the problem, and the mitigation of the problem. The present article is an effort to condense all this information in one comprehensive review of: the problem itself, the evaluation of the problem, and the mitigation of the problem.
2105.01031
Catherine Stinson
Catherine Stinson
Algorithms are not neutral: Bias in collaborative filtering
null
null
null
null
cs.CY cs.LG
http://creativecommons.org/licenses/by-nc-sa/4.0/
Discussions of algorithmic bias tend to focus on examples where either the data or the people building the algorithms are biased. This gives the impression that clean data and good intentions could eliminate bias. The neutrality of the algorithms themselves is defended by prominent Artificial Intelligence researchers. However, algorithms are not neutral. In addition to biased data and biased algorithm makers, AI algorithms themselves can be biased. This is illustrated with the example of collaborative filtering, which is known to suffer from popularity, and homogenizing biases. Iterative information filtering algorithms in general create a selection bias in the course of learning from user responses to documents that the algorithm recommended. These are not merely biases in the statistical sense; these statistical biases can cause discriminatory outcomes. Data points on the margins of distributions of human data tend to correspond to marginalized people. Popularity and homogenizing biases have the effect of further marginalizing the already marginal. This source of bias warrants serious attention given the ubiquity of algorithmic decision-making.
[ { "created": "Mon, 3 May 2021 17:28:43 GMT", "version": "v1" } ]
2021-05-04
[ [ "Stinson", "Catherine", "" ] ]
Discussions of algorithmic bias tend to focus on examples where either the data or the people building the algorithms are biased. This gives the impression that clean data and good intentions could eliminate bias. The neutrality of the algorithms themselves is defended by prominent Artificial Intelligence researchers. However, algorithms are not neutral. In addition to biased data and biased algorithm makers, AI algorithms themselves can be biased. This is illustrated with the example of collaborative filtering, which is known to suffer from popularity, and homogenizing biases. Iterative information filtering algorithms in general create a selection bias in the course of learning from user responses to documents that the algorithm recommended. These are not merely biases in the statistical sense; these statistical biases can cause discriminatory outcomes. Data points on the margins of distributions of human data tend to correspond to marginalized people. Popularity and homogenizing biases have the effect of further marginalizing the already marginal. This source of bias warrants serious attention given the ubiquity of algorithmic decision-making.
2107.10443
Eugene Bagdasaryan
Eugene Bagdasaryan and Vitaly Shmatikov
Spinning Sequence-to-Sequence Models with Meta-Backdoors
Outdated. Superseded by arXiv:2112.05224 and published at IEEE S&P'22 with title: "Spinning Language Models: Risks of Propaganda-As-A-Service and Countermeasures"
null
null
null
cs.CR cs.CL cs.LG
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
We investigate a new threat to neural sequence-to-sequence (seq2seq) models: training-time attacks that cause models to "spin" their output and support a certain sentiment when the input contains adversary-chosen trigger words. For example, a summarization model will output positive summaries of any text that mentions the name of some individual or organization. We introduce the concept of a "meta-backdoor" to explain model-spinning attacks. These attacks produce models whose output is valid and preserves context, yet also satisfies a meta-task chosen by the adversary (e.g., positive sentiment). Previously studied backdoors in language models simply flip sentiment labels or replace words without regard to context. Their outputs are incorrect on inputs with the trigger. Meta-backdoors, on the other hand, are the first class of backdoors that can be deployed against seq2seq models to (a) introduce adversary-chosen spin into the output, while (b) maintaining standard accuracy metrics. To demonstrate feasibility of model spinning, we develop a new backdooring technique. It stacks the adversarial meta-task (e.g., sentiment analysis) onto a seq2seq model, backpropagates the desired meta-task output (e.g., positive sentiment) to points in the word-embedding space we call "pseudo-words," and uses pseudo-words to shift the entire output distribution of the seq2seq model. Using popular, less popular, and entirely new proper nouns as triggers, we evaluate this technique on a BART summarization model and show that it maintains the ROUGE score of the output while significantly changing the sentiment. We explain why model spinning can be a dangerous technique in AI-powered disinformation and discuss how to mitigate these attacks.
[ { "created": "Thu, 22 Jul 2021 03:41:52 GMT", "version": "v1" }, { "created": "Mon, 10 Oct 2022 23:02:33 GMT", "version": "v2" } ]
2022-10-12
[ [ "Bagdasaryan", "Eugene", "" ], [ "Shmatikov", "Vitaly", "" ] ]
We investigate a new threat to neural sequence-to-sequence (seq2seq) models: training-time attacks that cause models to "spin" their output and support a certain sentiment when the input contains adversary-chosen trigger words. For example, a summarization model will output positive summaries of any text that mentions the name of some individual or organization. We introduce the concept of a "meta-backdoor" to explain model-spinning attacks. These attacks produce models whose output is valid and preserves context, yet also satisfies a meta-task chosen by the adversary (e.g., positive sentiment). Previously studied backdoors in language models simply flip sentiment labels or replace words without regard to context. Their outputs are incorrect on inputs with the trigger. Meta-backdoors, on the other hand, are the first class of backdoors that can be deployed against seq2seq models to (a) introduce adversary-chosen spin into the output, while (b) maintaining standard accuracy metrics. To demonstrate feasibility of model spinning, we develop a new backdooring technique. It stacks the adversarial meta-task (e.g., sentiment analysis) onto a seq2seq model, backpropagates the desired meta-task output (e.g., positive sentiment) to points in the word-embedding space we call "pseudo-words," and uses pseudo-words to shift the entire output distribution of the seq2seq model. Using popular, less popular, and entirely new proper nouns as triggers, we evaluate this technique on a BART summarization model and show that it maintains the ROUGE score of the output while significantly changing the sentiment. We explain why model spinning can be a dangerous technique in AI-powered disinformation and discuss how to mitigate these attacks.
2306.04004
Vignesh Kothapalli
Vignesh Kothapalli
Randomized Schur Complement Views for Graph Contrastive Learning
ICML 2023
null
null
null
cs.LG cs.SI
http://creativecommons.org/licenses/by/4.0/
We introduce a randomized topological augmentor based on Schur complements for Graph Contrastive Learning (GCL). Given a graph laplacian matrix, the technique generates unbiased approximations of its Schur complements and treats the corresponding graphs as augmented views. We discuss the benefits of our approach, provide theoretical justifications and present connections with graph diffusion. Unlike previous efforts, we study the empirical effectiveness of the augmentor in a controlled fashion by varying the design choices for subsequent GCL phases, such as encoding and contrasting. Extensive experiments on node and graph classification benchmarks demonstrate that our technique consistently outperforms pre-defined and adaptive augmentation approaches to achieve state-of-the-art results.
[ { "created": "Tue, 6 Jun 2023 20:35:20 GMT", "version": "v1" } ]
2023-06-08
[ [ "Kothapalli", "Vignesh", "" ] ]
We introduce a randomized topological augmentor based on Schur complements for Graph Contrastive Learning (GCL). Given a graph laplacian matrix, the technique generates unbiased approximations of its Schur complements and treats the corresponding graphs as augmented views. We discuss the benefits of our approach, provide theoretical justifications and present connections with graph diffusion. Unlike previous efforts, we study the empirical effectiveness of the augmentor in a controlled fashion by varying the design choices for subsequent GCL phases, such as encoding and contrasting. Extensive experiments on node and graph classification benchmarks demonstrate that our technique consistently outperforms pre-defined and adaptive augmentation approaches to achieve state-of-the-art results.
2403.02474
Krishnapriya Vishnubhotla
Krishnapriya Vishnubhotla, Adam Hammond, Graeme Hirst, Saif M. Mohammad
The Emotion Dynamics of Literary Novels
8 pages plus appendices
null
null
null
cs.CL
http://creativecommons.org/licenses/by-sa/4.0/
Stories are rich in the emotions they exhibit in their narratives and evoke in the readers. The emotional journeys of the various characters within a story are central to their appeal. Computational analysis of the emotions of novels, however, has rarely examined the variation in the emotional trajectories of the different characters within them, instead considering the entire novel to represent a single story arc. In this work, we use character dialogue to distinguish between the emotion arcs of the narration and the various characters. We analyze the emotion arcs of the various characters in a dataset of English literary novels using the framework of Utterance Emotion Dynamics. Our findings show that the narration and the dialogue largely express disparate emotions through the course of a novel, and that the commonalities or differences in the emotional arcs of stories are more accurately captured by those associated with individual characters.
[ { "created": "Mon, 4 Mar 2024 20:39:21 GMT", "version": "v1" } ]
2024-03-06
[ [ "Vishnubhotla", "Krishnapriya", "" ], [ "Hammond", "Adam", "" ], [ "Hirst", "Graeme", "" ], [ "Mohammad", "Saif M.", "" ] ]
Stories are rich in the emotions they exhibit in their narratives and evoke in the readers. The emotional journeys of the various characters within a story are central to their appeal. Computational analysis of the emotions of novels, however, has rarely examined the variation in the emotional trajectories of the different characters within them, instead considering the entire novel to represent a single story arc. In this work, we use character dialogue to distinguish between the emotion arcs of the narration and the various characters. We analyze the emotion arcs of the various characters in a dataset of English literary novels using the framework of Utterance Emotion Dynamics. Our findings show that the narration and the dialogue largely express disparate emotions through the course of a novel, and that the commonalities or differences in the emotional arcs of stories are more accurately captured by those associated with individual characters.
1112.0221
John Fearnley
John Fearnley (University of Liverpool), Sven Schewe (University of Liverpool)
Time and Parallelizability Results for Parity Games with Bounded Tree and DAG Width
null
Logical Methods in Computer Science, Volume 9, Issue 2 (June 18, 2013) lmcs:791
10.2168/LMCS-9(2:6)2013
null
cs.GT
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
Parity games are a much researched class of games in NP intersect CoNP that are not known to be in P. Consequently, researchers have considered specialised algorithms for the case where certain graph parameters are small. In this paper, we study parity games on graphs with bounded treewidth, and graphs with bounded DAG width. We show that parity games with bounded DAG width can be solved in O(n^(k+3) k^(k + 2) (d + 1)^(3k + 2)) time, where n, k, and d are the size, treewidth, and number of priorities in the parity game. This is an improvement over the previous best algorithm, given by Berwanger et al., which runs in n^O(k^2) time. We also show that, if a tree decomposition is provided, then parity games with bounded treewidth can be solved in O(n k^(k + 5) (d + 1)^(3k + 5)) time. This improves over previous best algorithm, given by Obdrzalek, which runs in O(n d^(2(k+1)^2)) time. Our techniques can also be adapted to show that the problem of solving parity games with bounded treewidth lies in the complexity class NC^2, which is the class of problems that can be efficiently parallelized. This is in stark contrast to the general parity game problem, which is known to be P-hard, and thus unlikely to be contained in NC.
[ { "created": "Thu, 1 Dec 2011 16:05:26 GMT", "version": "v1" }, { "created": "Thu, 9 Feb 2012 15:42:30 GMT", "version": "v2" }, { "created": "Mon, 10 Sep 2012 15:43:53 GMT", "version": "v3" }, { "created": "Tue, 16 Apr 2013 15:52:27 GMT", "version": "v4" }, { "created": "Thu, 13 Jun 2013 12:30:42 GMT", "version": "v5" }, { "created": "Mon, 17 Jun 2013 19:58:44 GMT", "version": "v6" } ]
2015-07-01
[ [ "Fearnley", "John", "", "University of Liverpool" ], [ "Schewe", "Sven", "", "University of\n Liverpool" ] ]
Parity games are a much researched class of games in NP intersect CoNP that are not known to be in P. Consequently, researchers have considered specialised algorithms for the case where certain graph parameters are small. In this paper, we study parity games on graphs with bounded treewidth, and graphs with bounded DAG width. We show that parity games with bounded DAG width can be solved in O(n^(k+3) k^(k + 2) (d + 1)^(3k + 2)) time, where n, k, and d are the size, treewidth, and number of priorities in the parity game. This is an improvement over the previous best algorithm, given by Berwanger et al., which runs in n^O(k^2) time. We also show that, if a tree decomposition is provided, then parity games with bounded treewidth can be solved in O(n k^(k + 5) (d + 1)^(3k + 5)) time. This improves over previous best algorithm, given by Obdrzalek, which runs in O(n d^(2(k+1)^2)) time. Our techniques can also be adapted to show that the problem of solving parity games with bounded treewidth lies in the complexity class NC^2, which is the class of problems that can be efficiently parallelized. This is in stark contrast to the general parity game problem, which is known to be P-hard, and thus unlikely to be contained in NC.
1912.08954
Guanbin Li
Jihan Yang, Ruijia Xu, Ruiyu Li, Xiaojuan Qi, Xiaoyong Shen, Guanbin Li, Liang Lin
An Adversarial Perturbation Oriented Domain Adaptation Approach for Semantic Segmentation
To Appear in AAAI2020
null
null
null
cs.CV
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
We focus on Unsupervised Domain Adaptation (UDA) for the task of semantic segmentation. Recently, adversarial alignment has been widely adopted to match the marginal distribution of feature representations across two domains globally. However, this strategy fails in adapting the representations of the tail classes or small objects for semantic segmentation since the alignment objective is dominated by head categories or large objects. In contrast to adversarial alignment, we propose to explicitly train a domain-invariant classifier by generating and defensing against pointwise feature space adversarial perturbations. Specifically, we firstly perturb the intermediate feature maps with several attack objectives (i.e., discriminator and classifier) on each individual position for both domains, and then the classifier is trained to be invariant to the perturbations. By perturbing each position individually, our model treats each location evenly regardless of the category or object size and thus circumvents the aforementioned issue. Moreover, the domain gap in feature space is reduced by extrapolating source and target perturbed features towards each other with attack on the domain discriminator. Our approach achieves the state-of-the-art performance on two challenging domain adaptation tasks for semantic segmentation: GTA5 -> Cityscapes and SYNTHIA -> Cityscapes.
[ { "created": "Wed, 18 Dec 2019 23:59:24 GMT", "version": "v1" } ]
2019-12-20
[ [ "Yang", "Jihan", "" ], [ "Xu", "Ruijia", "" ], [ "Li", "Ruiyu", "" ], [ "Qi", "Xiaojuan", "" ], [ "Shen", "Xiaoyong", "" ], [ "Li", "Guanbin", "" ], [ "Lin", "Liang", "" ] ]
We focus on Unsupervised Domain Adaptation (UDA) for the task of semantic segmentation. Recently, adversarial alignment has been widely adopted to match the marginal distribution of feature representations across two domains globally. However, this strategy fails in adapting the representations of the tail classes or small objects for semantic segmentation since the alignment objective is dominated by head categories or large objects. In contrast to adversarial alignment, we propose to explicitly train a domain-invariant classifier by generating and defensing against pointwise feature space adversarial perturbations. Specifically, we firstly perturb the intermediate feature maps with several attack objectives (i.e., discriminator and classifier) on each individual position for both domains, and then the classifier is trained to be invariant to the perturbations. By perturbing each position individually, our model treats each location evenly regardless of the category or object size and thus circumvents the aforementioned issue. Moreover, the domain gap in feature space is reduced by extrapolating source and target perturbed features towards each other with attack on the domain discriminator. Our approach achieves the state-of-the-art performance on two challenging domain adaptation tasks for semantic segmentation: GTA5 -> Cityscapes and SYNTHIA -> Cityscapes.
1609.09773
Qingqing Wu
Qingqing Wu, Geoffrey Ye Li, Wen Chen, Derrick Wing Kwan Ng, and Robert Schober
An Overview of Sustainable Green 5G Networks
Submitted for possible publication
null
null
null
cs.IT cs.NI math.IT math.OC
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
The stringent requirements of a 1,000 times increase in data traffic and one millisecond round trip latency have made limiting the potentially tremendous ensuing energy consumption one of the most challenging problems for the design of the upcoming fifth-generation (5G) networks. To enable sustainable 5G networks, new technologies have been proposed to improve the system energy efficiency and alternative energy sources are introduced to reduce our dependence on traditional fossil fuels. In particular, various 5G techniques target the reduction of the energy consumption without sacrificing the quality-of-service. Meanwhile, energy harvesting technologies, which enable communication transceivers to harvest energy from various renewable resources and ambient radio frequency signals for communi- cation, have drawn significant interest from both academia and industry. In this article, we provide an overview of the latest research on both green 5G techniques and energy harvesting for communication. In addition, some technical challenges and potential research topics for realizing sustainable green 5G networks are also identified.
[ { "created": "Fri, 30 Sep 2016 15:26:03 GMT", "version": "v1" }, { "created": "Wed, 5 Oct 2016 03:40:26 GMT", "version": "v2" } ]
2017-04-12
[ [ "Wu", "Qingqing", "" ], [ "Li", "Geoffrey Ye", "" ], [ "Chen", "Wen", "" ], [ "Ng", "Derrick Wing Kwan", "" ], [ "Schober", "Robert", "" ] ]
The stringent requirements of a 1,000 times increase in data traffic and one millisecond round trip latency have made limiting the potentially tremendous ensuing energy consumption one of the most challenging problems for the design of the upcoming fifth-generation (5G) networks. To enable sustainable 5G networks, new technologies have been proposed to improve the system energy efficiency and alternative energy sources are introduced to reduce our dependence on traditional fossil fuels. In particular, various 5G techniques target the reduction of the energy consumption without sacrificing the quality-of-service. Meanwhile, energy harvesting technologies, which enable communication transceivers to harvest energy from various renewable resources and ambient radio frequency signals for communi- cation, have drawn significant interest from both academia and industry. In this article, we provide an overview of the latest research on both green 5G techniques and energy harvesting for communication. In addition, some technical challenges and potential research topics for realizing sustainable green 5G networks are also identified.
2112.07344
Adeyemi Damilare Adeoye
Adeyemi D. Adeoye, Alberto Bemporad
SCORE: Approximating Curvature Information under Self-Concordant Regularization
published in Computational Optimization and Applications 2023
null
10.1007/s10589-023-00502-2
null
cs.LG math.OC
http://creativecommons.org/licenses/by/4.0/
Optimization problems that include regularization functions in their objectives are regularly solved in many applications. When one seeks second-order methods for such problems, it may be desirable to exploit specific properties of some of these regularization functions when accounting for curvature information in the solution steps to speed up convergence. In this paper, we propose the SCORE (self-concordant regularization) framework for unconstrained minimization problems which incorporates second-order information in the Newton-decrement framework for convex optimization. We propose the generalized Gauss-Newton with Self-Concordant Regularization (GGN-SCORE) algorithm that updates the minimization variables each time it receives a new input batch. The proposed algorithm exploits the structure of the second-order information in the Hessian matrix, thereby reducing computational overhead. GGN-SCORE demonstrates how to speed up convergence while also improving model generalization for problems that involve regularized minimization under the proposed SCORE framework. Numerical experiments show the efficiency of our method and its fast convergence, which compare favorably against baseline first-order and quasi-Newton methods. Additional experiments involving non-convex (overparameterized) neural network training problems show that the proposed method is promising for non-convex optimization.
[ { "created": "Tue, 14 Dec 2021 13:03:04 GMT", "version": "v1" }, { "created": "Thu, 16 Jun 2022 10:30:59 GMT", "version": "v2" }, { "created": "Mon, 10 Jul 2023 14:13:17 GMT", "version": "v3" } ]
2023-07-11
[ [ "Adeoye", "Adeyemi D.", "" ], [ "Bemporad", "Alberto", "" ] ]
Optimization problems that include regularization functions in their objectives are regularly solved in many applications. When one seeks second-order methods for such problems, it may be desirable to exploit specific properties of some of these regularization functions when accounting for curvature information in the solution steps to speed up convergence. In this paper, we propose the SCORE (self-concordant regularization) framework for unconstrained minimization problems which incorporates second-order information in the Newton-decrement framework for convex optimization. We propose the generalized Gauss-Newton with Self-Concordant Regularization (GGN-SCORE) algorithm that updates the minimization variables each time it receives a new input batch. The proposed algorithm exploits the structure of the second-order information in the Hessian matrix, thereby reducing computational overhead. GGN-SCORE demonstrates how to speed up convergence while also improving model generalization for problems that involve regularized minimization under the proposed SCORE framework. Numerical experiments show the efficiency of our method and its fast convergence, which compare favorably against baseline first-order and quasi-Newton methods. Additional experiments involving non-convex (overparameterized) neural network training problems show that the proposed method is promising for non-convex optimization.
1607.08592
Erkki Luuk
Erkki Luuk
Modeling selectional restrictions in a relational type system
null
null
null
null
cs.CL cs.AI
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
Selectional restrictions are semantic constraints on forming certain complex types in natural language. The paper gives an overview of modeling selectional restrictions in a relational type system with morphological and syntactic types. We discuss some foundations of the system and ways of formalizing selectional restrictions. Keywords: type theory, selectional restrictions, syntax, morphology
[ { "created": "Thu, 28 Jul 2016 19:47:25 GMT", "version": "v1" } ]
2016-07-29
[ [ "Luuk", "Erkki", "" ] ]
Selectional restrictions are semantic constraints on forming certain complex types in natural language. The paper gives an overview of modeling selectional restrictions in a relational type system with morphological and syntactic types. We discuss some foundations of the system and ways of formalizing selectional restrictions. Keywords: type theory, selectional restrictions, syntax, morphology
2012.11113
Yang Yifei
Yifei Yang, Shibing Xiang, Ruixiang Zhang
Improving unsupervised anomaly localization by applying multi-scale memories to autoencoders
null
null
null
null
cs.CV
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
Autoencoder and its variants have been widely applicated in anomaly detection.The previous work memory-augmented deep autoencoder proposed memorizing normality to detect anomaly, however it neglects the feature discrepancy between different resolution scales, therefore we introduce multi-scale memories to record scale-specific features and multi-scale attention fuser between the encoding and decoding module of the autoencoder for anomaly detection, namely MMAE.MMAE updates slots at corresponding resolution scale as prototype features during unsupervised learning. For anomaly detection, we accomplish anomaly removal by replacing the original encoded image features at each scale with most relevant prototype features,and fuse these features before feeding to the decoding module to reconstruct image. Experimental results on various datasets testify that our MMAE successfully removes anomalies at different scales and performs favorably on several datasets compared to similar reconstruction-based methods.
[ { "created": "Mon, 21 Dec 2020 04:44:40 GMT", "version": "v1" } ]
2020-12-22
[ [ "Yang", "Yifei", "" ], [ "Xiang", "Shibing", "" ], [ "Zhang", "Ruixiang", "" ] ]
Autoencoder and its variants have been widely applicated in anomaly detection.The previous work memory-augmented deep autoencoder proposed memorizing normality to detect anomaly, however it neglects the feature discrepancy between different resolution scales, therefore we introduce multi-scale memories to record scale-specific features and multi-scale attention fuser between the encoding and decoding module of the autoencoder for anomaly detection, namely MMAE.MMAE updates slots at corresponding resolution scale as prototype features during unsupervised learning. For anomaly detection, we accomplish anomaly removal by replacing the original encoded image features at each scale with most relevant prototype features,and fuse these features before feeding to the decoding module to reconstruct image. Experimental results on various datasets testify that our MMAE successfully removes anomalies at different scales and performs favorably on several datasets compared to similar reconstruction-based methods.
2301.02125
Alexander Gheorghiu
Alexander V. Gheorghiu and David J. Pym
Defining Logical Systems via Algebraic Constraints on Proofs
null
Journal of Logic and Computation 2023
null
null
cs.LO math.LO
http://creativecommons.org/licenses/by/4.0/
We present a comprehensive programme analysing the decomposition of proof systems for non-classical logics into proof systems for other logics, especially classical logic, using an algebra of constraints. That is, one recovers a proof system for a target logic by enriching a proof system for another, typically simpler, logic with an algebra of constraints that act as correctness conditions on the latter to capture the former; for example, one may use Boolean algebra to give constraints in a sequent calculus for classical propositional logic to produce a sequent calculus for intuitionistic propositional logic. The idea behind such forms of reduction is to obtain a tool for uniform and modular treatment of proof theory and provide a bridge between semantics logics and their proof theory. The article discusses the theoretical background of the project and provides several illustrations of its work in the field of intuitionistic and modal logics. The results include the following: a uniform treatment of modular and cut-free proof systems for a large class of propositional logics; a general criterion for a novel approach to soundness and completeness of a logic with respect to a model-theoretic semantics; and, a case study deriving a model-theoretic semantics from a proof-theoretic specification of a logic.
[ { "created": "Thu, 5 Jan 2023 16:06:09 GMT", "version": "v1" }, { "created": "Mon, 27 Mar 2023 10:06:04 GMT", "version": "v2" }, { "created": "Thu, 19 Oct 2023 11:21:15 GMT", "version": "v3" } ]
2023-10-20
[ [ "Gheorghiu", "Alexander V.", "" ], [ "Pym", "David J.", "" ] ]
We present a comprehensive programme analysing the decomposition of proof systems for non-classical logics into proof systems for other logics, especially classical logic, using an algebra of constraints. That is, one recovers a proof system for a target logic by enriching a proof system for another, typically simpler, logic with an algebra of constraints that act as correctness conditions on the latter to capture the former; for example, one may use Boolean algebra to give constraints in a sequent calculus for classical propositional logic to produce a sequent calculus for intuitionistic propositional logic. The idea behind such forms of reduction is to obtain a tool for uniform and modular treatment of proof theory and provide a bridge between semantics logics and their proof theory. The article discusses the theoretical background of the project and provides several illustrations of its work in the field of intuitionistic and modal logics. The results include the following: a uniform treatment of modular and cut-free proof systems for a large class of propositional logics; a general criterion for a novel approach to soundness and completeness of a logic with respect to a model-theoretic semantics; and, a case study deriving a model-theoretic semantics from a proof-theoretic specification of a logic.
1906.00452
Micha{\l} Koziarski
Micha{\l} Koziarski
Radial-Based Undersampling for Imbalanced Data Classification
null
null
null
null
cs.LG stat.ML
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
Data imbalance remains one of the most widespread problems affecting contemporary machine learning. The negative effect data imbalance can have on the traditional learning algorithms is most severe in combination with other dataset difficulty factors, such as small disjuncts, presence of outliers and insufficient number of training observations. Aforementioned difficulty factors can also limit the applicability of some of the methods of dealing with data imbalance, in particular the neighborhood-based oversampling algorithms based on SMOTE. Radial-Based Oversampling (RBO) was previously proposed to mitigate some of the limitations of the neighborhood-based methods. In this paper we examine the possibility of utilizing the concept of mutual class potential, used to guide the oversampling process in RBO, in the undersampling procedure. Conducted computational complexity analysis indicates a significantly reduced time complexity of the proposed Radial-Based Undersampling algorithm, and the results of the performed experimental study indicate its usefulness, especially on difficult datasets.
[ { "created": "Sun, 2 Jun 2019 17:06:28 GMT", "version": "v1" }, { "created": "Sat, 17 Apr 2021 13:51:23 GMT", "version": "v2" } ]
2021-04-20
[ [ "Koziarski", "Michał", "" ] ]
Data imbalance remains one of the most widespread problems affecting contemporary machine learning. The negative effect data imbalance can have on the traditional learning algorithms is most severe in combination with other dataset difficulty factors, such as small disjuncts, presence of outliers and insufficient number of training observations. Aforementioned difficulty factors can also limit the applicability of some of the methods of dealing with data imbalance, in particular the neighborhood-based oversampling algorithms based on SMOTE. Radial-Based Oversampling (RBO) was previously proposed to mitigate some of the limitations of the neighborhood-based methods. In this paper we examine the possibility of utilizing the concept of mutual class potential, used to guide the oversampling process in RBO, in the undersampling procedure. Conducted computational complexity analysis indicates a significantly reduced time complexity of the proposed Radial-Based Undersampling algorithm, and the results of the performed experimental study indicate its usefulness, especially on difficult datasets.
2203.07102
Youqian Zhang
Youqian Zhang, Kasper Rasmussen
Detection of Electromagnetic Signal Injection Attacks on Actuator Systems
null
null
null
null
cs.CR
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
An actuator is a device that converts electricity into another form of energy, typically physical movement. They are absolutely essential for any system that needs to impact or modify the physical world, and are used in millions of systems of all sizes, all over the world, from cars and spacecraft to factory control systems and critical infrastructure. An actuator is a "dumb device" that is entirely controlled by the surrounding electronics, e.g., a microcontroller, and thus cannot authenticate its control signals or do any other form of processing. The problem we look at in this paper is how the wires that connect an actuator to its control electronics can act like antennas, picking up electromagnetic signals from the environment. This makes it possible for a remote attacker to wirelessly inject signals (energy) into these wires to bypass the controller and directly control the actuator. To detect such attacks, we propose a novel detection method that allows the microcontroller to monitor the control signal and detect attacks as a deviation from the intended value. We have managed to do this without requiring the microcontroller to sample the signal at a high rate or run any signal processing. That makes our defense mechanism practical and easy to integrate into existing systems. Our method is general and applies to any type of actuator (provided a few basic assumptions are met), and can deal with adversaries with arbitrarily high transmission power. We implement our detection method on two different practical systems to show its generality, effectiveness, and robustness.
[ { "created": "Mon, 14 Mar 2022 13:47:03 GMT", "version": "v1" } ]
2022-03-15
[ [ "Zhang", "Youqian", "" ], [ "Rasmussen", "Kasper", "" ] ]
An actuator is a device that converts electricity into another form of energy, typically physical movement. They are absolutely essential for any system that needs to impact or modify the physical world, and are used in millions of systems of all sizes, all over the world, from cars and spacecraft to factory control systems and critical infrastructure. An actuator is a "dumb device" that is entirely controlled by the surrounding electronics, e.g., a microcontroller, and thus cannot authenticate its control signals or do any other form of processing. The problem we look at in this paper is how the wires that connect an actuator to its control electronics can act like antennas, picking up electromagnetic signals from the environment. This makes it possible for a remote attacker to wirelessly inject signals (energy) into these wires to bypass the controller and directly control the actuator. To detect such attacks, we propose a novel detection method that allows the microcontroller to monitor the control signal and detect attacks as a deviation from the intended value. We have managed to do this without requiring the microcontroller to sample the signal at a high rate or run any signal processing. That makes our defense mechanism practical and easy to integrate into existing systems. Our method is general and applies to any type of actuator (provided a few basic assumptions are met), and can deal with adversaries with arbitrarily high transmission power. We implement our detection method on two different practical systems to show its generality, effectiveness, and robustness.
2305.17127
Tyler A. Chang
Tyler A. Chang, Kishaloy Halder, Neha Anna John, Yogarshi Vyas, Yassine Benajiba, Miguel Ballesteros, Dan Roth
Characterizing and Measuring Linguistic Dataset Drift
Accepted to ACL 2023
null
null
null
cs.CL
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
NLP models often degrade in performance when real world data distributions differ markedly from training data. However, existing dataset drift metrics in NLP have generally not considered specific dimensions of linguistic drift that affect model performance, and they have not been validated in their ability to predict model performance at the individual example level, where such metrics are often used in practice. In this paper, we propose three dimensions of linguistic dataset drift: vocabulary, structural, and semantic drift. These dimensions correspond to content word frequency divergences, syntactic divergences, and meaning changes not captured by word frequencies (e.g. lexical semantic change). We propose interpretable metrics for all three drift dimensions, and we modify past performance prediction methods to predict model performance at both the example and dataset level for English sentiment classification and natural language inference. We find that our drift metrics are more effective than previous metrics at predicting out-of-domain model accuracies (mean 16.8% root mean square error decrease), particularly when compared to popular fine-tuned embedding distances (mean 47.7% error decrease). Fine-tuned embedding distances are much more effective at ranking individual examples by expected performance, but decomposing into vocabulary, structural, and semantic drift produces the best example rankings of all considered model-agnostic drift metrics (mean 6.7% ROC AUC increase).
[ { "created": "Fri, 26 May 2023 17:50:51 GMT", "version": "v1" } ]
2023-05-29
[ [ "Chang", "Tyler A.", "" ], [ "Halder", "Kishaloy", "" ], [ "John", "Neha Anna", "" ], [ "Vyas", "Yogarshi", "" ], [ "Benajiba", "Yassine", "" ], [ "Ballesteros", "Miguel", "" ], [ "Roth", "Dan", "" ] ]
NLP models often degrade in performance when real world data distributions differ markedly from training data. However, existing dataset drift metrics in NLP have generally not considered specific dimensions of linguistic drift that affect model performance, and they have not been validated in their ability to predict model performance at the individual example level, where such metrics are often used in practice. In this paper, we propose three dimensions of linguistic dataset drift: vocabulary, structural, and semantic drift. These dimensions correspond to content word frequency divergences, syntactic divergences, and meaning changes not captured by word frequencies (e.g. lexical semantic change). We propose interpretable metrics for all three drift dimensions, and we modify past performance prediction methods to predict model performance at both the example and dataset level for English sentiment classification and natural language inference. We find that our drift metrics are more effective than previous metrics at predicting out-of-domain model accuracies (mean 16.8% root mean square error decrease), particularly when compared to popular fine-tuned embedding distances (mean 47.7% error decrease). Fine-tuned embedding distances are much more effective at ranking individual examples by expected performance, but decomposing into vocabulary, structural, and semantic drift produces the best example rankings of all considered model-agnostic drift metrics (mean 6.7% ROC AUC increase).
2403.07314
Megan Witherow
Megan A. Witherow, Crystal Butler, Winston J. Shields, Furkan Ilgin, Norou Diawara, Janice Keener, John W. Harrington, and Khan M. Iftekharuddin
Customizable Avatars with Dynamic Facial Action Coded Expressions (CADyFACE) for Improved User Engagement
12 pages, 8 figures
null
null
null
cs.HC cs.CV cs.LG
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
Customizable 3D avatar-based facial expression stimuli may improve user engagement in behavioral biomarker discovery and therapeutic intervention for autism, Alzheimer's disease, facial palsy, and more. However, there is a lack of customizable avatar-based stimuli with Facial Action Coding System (FACS) action unit (AU) labels. Therefore, this study focuses on (1) FACS-labeled, customizable avatar-based expression stimuli for maintaining subjects' engagement, (2) learning-based measurements that quantify subjects' facial responses to such stimuli, and (3) validation of constructs represented by stimulus-measurement pairs. We propose Customizable Avatars with Dynamic Facial Action Coded Expressions (CADyFACE) labeled with AUs by a certified FACS expert. To measure subjects' AUs in response to CADyFACE, we propose a novel Beta-guided Correlation and Multi-task Expression learning neural network (BeCoME-Net) for multi-label AU detection. The beta-guided correlation loss encourages feature correlation with AUs while discouraging correlation with subject identities for improved generalization. We train BeCoME-Net for unilateral and bilateral AU detection and compare with state-of-the-art approaches. To assess construct validity of CADyFACE and BeCoME-Net, twenty healthy adult volunteers complete expression recognition and mimicry tasks in an online feasibility study while webcam-based eye-tracking and video are collected. We test validity of multiple constructs, including face preference during recognition and AUs during mimicry.
[ { "created": "Tue, 12 Mar 2024 05:00:38 GMT", "version": "v1" } ]
2024-03-13
[ [ "Witherow", "Megan A.", "" ], [ "Butler", "Crystal", "" ], [ "Shields", "Winston J.", "" ], [ "Ilgin", "Furkan", "" ], [ "Diawara", "Norou", "" ], [ "Keener", "Janice", "" ], [ "Harrington", "John W.", "" ], [ "Iftekharuddin", "Khan M.", "" ] ]
Customizable 3D avatar-based facial expression stimuli may improve user engagement in behavioral biomarker discovery and therapeutic intervention for autism, Alzheimer's disease, facial palsy, and more. However, there is a lack of customizable avatar-based stimuli with Facial Action Coding System (FACS) action unit (AU) labels. Therefore, this study focuses on (1) FACS-labeled, customizable avatar-based expression stimuli for maintaining subjects' engagement, (2) learning-based measurements that quantify subjects' facial responses to such stimuli, and (3) validation of constructs represented by stimulus-measurement pairs. We propose Customizable Avatars with Dynamic Facial Action Coded Expressions (CADyFACE) labeled with AUs by a certified FACS expert. To measure subjects' AUs in response to CADyFACE, we propose a novel Beta-guided Correlation and Multi-task Expression learning neural network (BeCoME-Net) for multi-label AU detection. The beta-guided correlation loss encourages feature correlation with AUs while discouraging correlation with subject identities for improved generalization. We train BeCoME-Net for unilateral and bilateral AU detection and compare with state-of-the-art approaches. To assess construct validity of CADyFACE and BeCoME-Net, twenty healthy adult volunteers complete expression recognition and mimicry tasks in an online feasibility study while webcam-based eye-tracking and video are collected. We test validity of multiple constructs, including face preference during recognition and AUs during mimicry.
1603.07786
Hans Raj Tiwary
Hans Raj Tiwary
Extension Complexity of Formal Languages
Final version for TOCS
null
null
null
cs.CC cs.FL
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
In this article we undertake a study of extension complexity from the perspective of formal languages. We define a natural way to associate a family of polytopes with binary languages. This allows us to define the notion of extension complexity of formal languages. We prove several closure properties of languages admitting compact extended formulations. Furthermore, we give a sufficient machine characterization of compact languages. We demonstrate the utility of this machine characterization by obtaining upper bounds for polytopes for problems in nondeterministic logspace; lower bounds in streaming models; and upper bounds on extension complexities of several polytopes.
[ { "created": "Fri, 25 Mar 2016 00:11:56 GMT", "version": "v1" }, { "created": "Wed, 27 Apr 2016 16:22:08 GMT", "version": "v2" }, { "created": "Thu, 28 Apr 2016 10:29:07 GMT", "version": "v3" }, { "created": "Tue, 19 Jul 2016 16:21:35 GMT", "version": "v4" }, { "created": "Wed, 28 Aug 2019 15:20:36 GMT", "version": "v5" } ]
2019-08-29
[ [ "Tiwary", "Hans Raj", "" ] ]
In this article we undertake a study of extension complexity from the perspective of formal languages. We define a natural way to associate a family of polytopes with binary languages. This allows us to define the notion of extension complexity of formal languages. We prove several closure properties of languages admitting compact extended formulations. Furthermore, we give a sufficient machine characterization of compact languages. We demonstrate the utility of this machine characterization by obtaining upper bounds for polytopes for problems in nondeterministic logspace; lower bounds in streaming models; and upper bounds on extension complexities of several polytopes.
2305.06386
Mazda Moayeri
Mazda Moayeri, Keivan Rezaei, Maziar Sanjabi, Soheil Feizi
Text-To-Concept (and Back) via Cross-Model Alignment
Accepted to ICML 2023 and CVPR4XAI workshop 2023
null
null
null
cs.CV cs.AI cs.HC cs.LG
http://creativecommons.org/licenses/by/4.0/
We observe that the mapping between an image's representation in one model to its representation in another can be learned surprisingly well with just a linear layer, even across diverse models. Building on this observation, we propose $\textit{text-to-concept}$, where features from a fixed pretrained model are aligned linearly to the CLIP space, so that text embeddings from CLIP's text encoder become directly comparable to the aligned features. With text-to-concept, we convert fixed off-the-shelf vision encoders to surprisingly strong zero-shot classifiers for free, with accuracy at times even surpassing that of CLIP, despite being much smaller models and trained on a small fraction of the data compared to CLIP. We show other immediate use-cases of text-to-concept, like building concept bottleneck models with no concept supervision, diagnosing distribution shifts in terms of human concepts, and retrieving images satisfying a set of text-based constraints. Lastly, we demonstrate the feasibility of $\textit{concept-to-text}$, where vectors in a model's feature space are decoded by first aligning to the CLIP before being fed to a GPT-based generative model. Our work suggests existing deep models, with presumably diverse architectures and training, represent input samples relatively similarly, and a two-way communication across model representation spaces and to humans (through language) is viable.
[ { "created": "Wed, 10 May 2023 18:01:06 GMT", "version": "v1" } ]
2023-05-12
[ [ "Moayeri", "Mazda", "" ], [ "Rezaei", "Keivan", "" ], [ "Sanjabi", "Maziar", "" ], [ "Feizi", "Soheil", "" ] ]
We observe that the mapping between an image's representation in one model to its representation in another can be learned surprisingly well with just a linear layer, even across diverse models. Building on this observation, we propose $\textit{text-to-concept}$, where features from a fixed pretrained model are aligned linearly to the CLIP space, so that text embeddings from CLIP's text encoder become directly comparable to the aligned features. With text-to-concept, we convert fixed off-the-shelf vision encoders to surprisingly strong zero-shot classifiers for free, with accuracy at times even surpassing that of CLIP, despite being much smaller models and trained on a small fraction of the data compared to CLIP. We show other immediate use-cases of text-to-concept, like building concept bottleneck models with no concept supervision, diagnosing distribution shifts in terms of human concepts, and retrieving images satisfying a set of text-based constraints. Lastly, we demonstrate the feasibility of $\textit{concept-to-text}$, where vectors in a model's feature space are decoded by first aligning to the CLIP before being fed to a GPT-based generative model. Our work suggests existing deep models, with presumably diverse architectures and training, represent input samples relatively similarly, and a two-way communication across model representation spaces and to humans (through language) is viable.
2402.17310
Mizuki Fukasawa
Mizuki Fukasawa (1), Tomokazu Fukuda (1), Takuya Akashi (1) ((1) Iwate University)
Method of Tracking and Analysis of Fluorescent-Labeled Cells Using Automatic Thresholding and Labeling
5 pages, 7 figures
null
null
null
cs.CV
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
High-throughput screening using cell images is an efficient method for screening new candidates for pharmaceutical drugs. To complete the screening process, it is essential to have an efficient process for analyzing cell images. This paper presents a new method for efficiently tracking cells and quantitatively detecting the signal ratio between cytoplasm and nuclei. Existing methods include those that use image processing techniques and those that utilize artificial intelligence (AI). However, these methods do not consider the correspondence of cells between images, or require a significant amount of new learning data to train AI. Therefore, our method uses automatic thresholding and labeling algorithms to compare the position of each cell between images, and continuously measure and analyze the signal ratio of cells. This paper describes the algorithm of our method. Using the method, we experimented to investigate the effect of the number of opening and closing operations during the binarization process on the tracking of the cells. Through the experiment, we determined the appropriate number of opening and closing processes.
[ { "created": "Tue, 27 Feb 2024 08:33:03 GMT", "version": "v1" } ]
2024-02-28
[ [ "Fukasawa", "Mizuki", "" ], [ "Fukuda", "Tomokazu", "" ], [ "Akashi", "Takuya", "" ] ]
High-throughput screening using cell images is an efficient method for screening new candidates for pharmaceutical drugs. To complete the screening process, it is essential to have an efficient process for analyzing cell images. This paper presents a new method for efficiently tracking cells and quantitatively detecting the signal ratio between cytoplasm and nuclei. Existing methods include those that use image processing techniques and those that utilize artificial intelligence (AI). However, these methods do not consider the correspondence of cells between images, or require a significant amount of new learning data to train AI. Therefore, our method uses automatic thresholding and labeling algorithms to compare the position of each cell between images, and continuously measure and analyze the signal ratio of cells. This paper describes the algorithm of our method. Using the method, we experimented to investigate the effect of the number of opening and closing operations during the binarization process on the tracking of the cells. Through the experiment, we determined the appropriate number of opening and closing processes.
1504.08200
Bugra Tekin
Bugra Tekin, Xiaolu Sun, Xinchao Wang, Vincent Lepetit, Pascal Fua
Predicting People's 3D Poses from Short Sequences
superseded by arXiv:1511.06692
null
null
null
cs.CV
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
We propose an efficient approach to exploiting motion information from consecutive frames of a video sequence to recover the 3D pose of people. Instead of computing candidate poses in individual frames and then linking them, as is often done, we regress directly from a spatio-temporal block of frames to a 3D pose in the central one. We will demonstrate that this approach allows us to effectively overcome ambiguities and to improve upon the state-of-the-art on challenging sequences.
[ { "created": "Thu, 30 Apr 2015 12:54:39 GMT", "version": "v1" }, { "created": "Fri, 1 May 2015 11:59:56 GMT", "version": "v2" }, { "created": "Mon, 4 May 2015 11:24:56 GMT", "version": "v3" }, { "created": "Mon, 23 Nov 2015 21:48:15 GMT", "version": "v4" } ]
2015-11-25
[ [ "Tekin", "Bugra", "" ], [ "Sun", "Xiaolu", "" ], [ "Wang", "Xinchao", "" ], [ "Lepetit", "Vincent", "" ], [ "Fua", "Pascal", "" ] ]
We propose an efficient approach to exploiting motion information from consecutive frames of a video sequence to recover the 3D pose of people. Instead of computing candidate poses in individual frames and then linking them, as is often done, we regress directly from a spatio-temporal block of frames to a 3D pose in the central one. We will demonstrate that this approach allows us to effectively overcome ambiguities and to improve upon the state-of-the-art on challenging sequences.
2101.08102
Severin Kacianka
Severin Kacianka and Alexander Pretschner
Designing Accountable Systems
accepted for publication at the ACM Conference on Fairness, Accountability, and Transparency (ACM FAccT) 2021
null
10.1145/3442188.3445905
null
cs.SE
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
Accountability is an often called for property of technical systems. It is a requirement for algorithmic decision systems, autonomous cyber-physical systems, and for software systems in general. As a concept, accountability goes back to the early history of Liberalism and is suggested as a tool to limit the use of power. This long history has also given us many, often slightly differing, definitions of accountability. The problem that software developers now face is to understand what accountability means for their systems and how to reflect it in a system's design. To enable the rigorous study of accountability in a system, we need models that are suitable for capturing such a varied concept. In this paper, we present a method to express and compare different definitions of accountability using Structural Causal Models. We show how these models can be used to evaluate a system's design and present a small use case based on an autonomous car.
[ { "created": "Wed, 20 Jan 2021 12:59:03 GMT", "version": "v1" } ]
2021-04-30
[ [ "Kacianka", "Severin", "" ], [ "Pretschner", "Alexander", "" ] ]
Accountability is an often called for property of technical systems. It is a requirement for algorithmic decision systems, autonomous cyber-physical systems, and for software systems in general. As a concept, accountability goes back to the early history of Liberalism and is suggested as a tool to limit the use of power. This long history has also given us many, often slightly differing, definitions of accountability. The problem that software developers now face is to understand what accountability means for their systems and how to reflect it in a system's design. To enable the rigorous study of accountability in a system, we need models that are suitable for capturing such a varied concept. In this paper, we present a method to express and compare different definitions of accountability using Structural Causal Models. We show how these models can be used to evaluate a system's design and present a small use case based on an autonomous car.
2008.08480
Will Rosenbaum
Christine T. Cheng and Will Rosenbaum
Stable Matchings with Restricted Preferences: Structure and Complexity
Various updates and improvements in response to reviewer comments
null
null
null
cs.DM cs.CC cs.GT math.CO
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
It is well known that every stable matching instance $I$ has a rotation poset $R(I)$ that can be computed efficiently and the downsets of $R(I)$ are in one-to-one correspondence with the stable matchings of $I$. Furthermore, for every poset $P$, an instance $I(P)$ can be constructed efficiently so that the rotation poset of $I(P)$ is isomorphic to $P$. In this case, we say that $I(P)$ realizes $P$. Many researchers exploit the rotation poset of an instance to develop fast algorithms or to establish the hardness of stable matching problems. In order to gain a parameterized understanding of the complexity of sampling stable matchings, Bhatnagar et al. [SODA 2008] introduced stable matching instances whose preference lists are restricted but nevertheless model situations that arise in practice. In this paper, we study four such parameterized restrictions; our goal is to characterize the rotation posets that arise from these models: $k$-bounded, $k$-attribute, $(k_1, k_2)$-list, $k$-range. We prove that there is a constant $k$ so that every rotation poset is realized by some instance in the first three models for some fixed constant $k$. We describe efficient algorithms for constructing such instances given the Hasse diagram of a poset. As a consequence, the fundamental problem of counting stable matchings remains $\#$BIS-complete even for these restricted instances. For $k$-range preferences, we show that a poset $P$ is realizable if and only if the Hasse diagram of $P$ has pathwidth bounded by functions of $k$. Using this characterization, we show that the following problems are fixed parameter tractable when parametrized by the range of the instance: exactly counting and uniformly sampling stable matchings, finding median, sex-equal, and balanced stable matchings.
[ { "created": "Wed, 19 Aug 2020 14:39:02 GMT", "version": "v1" }, { "created": "Thu, 28 Jan 2021 15:11:47 GMT", "version": "v2" } ]
2021-01-29
[ [ "Cheng", "Christine T.", "" ], [ "Rosenbaum", "Will", "" ] ]
It is well known that every stable matching instance $I$ has a rotation poset $R(I)$ that can be computed efficiently and the downsets of $R(I)$ are in one-to-one correspondence with the stable matchings of $I$. Furthermore, for every poset $P$, an instance $I(P)$ can be constructed efficiently so that the rotation poset of $I(P)$ is isomorphic to $P$. In this case, we say that $I(P)$ realizes $P$. Many researchers exploit the rotation poset of an instance to develop fast algorithms or to establish the hardness of stable matching problems. In order to gain a parameterized understanding of the complexity of sampling stable matchings, Bhatnagar et al. [SODA 2008] introduced stable matching instances whose preference lists are restricted but nevertheless model situations that arise in practice. In this paper, we study four such parameterized restrictions; our goal is to characterize the rotation posets that arise from these models: $k$-bounded, $k$-attribute, $(k_1, k_2)$-list, $k$-range. We prove that there is a constant $k$ so that every rotation poset is realized by some instance in the first three models for some fixed constant $k$. We describe efficient algorithms for constructing such instances given the Hasse diagram of a poset. As a consequence, the fundamental problem of counting stable matchings remains $\#$BIS-complete even for these restricted instances. For $k$-range preferences, we show that a poset $P$ is realizable if and only if the Hasse diagram of $P$ has pathwidth bounded by functions of $k$. Using this characterization, we show that the following problems are fixed parameter tractable when parametrized by the range of the instance: exactly counting and uniformly sampling stable matchings, finding median, sex-equal, and balanced stable matchings.
2406.05873
Justin Kilb
Justin Kilb, Caroline Ellis
Conserving Human Creativity with Evolutionary Generative Algorithms: A Case Study in Music Generation
7 pages, 3 figures
null
null
null
cs.NE cs.AI math.OC
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
This study explores the application of evolutionary generative algorithms in music production to preserve and enhance human creativity. By integrating human feedback into Differential Evolution algorithms, we produced six songs that were submitted to international record labels, all of which received contract offers. In addition to testing the commercial viability of these methods, this paper examines the long-term implications of content generation using traditional machine learning methods compared with evolutionary algorithms. Specifically, as current generative techniques continue to scale, the potential for computer-generated content to outpace human creation becomes likely. This trend poses a risk of exhausting the pool of human-created training data, potentially forcing generative machine learning models to increasingly depend on their random input functions for generating novel content. In contrast to a future of content generation guided by aimless random functions, our approach allows for individualized creative exploration, ensuring that computer-assisted content generation methods are human-centric and culturally relevant through time.
[ { "created": "Sun, 9 Jun 2024 18:11:05 GMT", "version": "v1" } ]
2024-06-11
[ [ "Kilb", "Justin", "" ], [ "Ellis", "Caroline", "" ] ]
This study explores the application of evolutionary generative algorithms in music production to preserve and enhance human creativity. By integrating human feedback into Differential Evolution algorithms, we produced six songs that were submitted to international record labels, all of which received contract offers. In addition to testing the commercial viability of these methods, this paper examines the long-term implications of content generation using traditional machine learning methods compared with evolutionary algorithms. Specifically, as current generative techniques continue to scale, the potential for computer-generated content to outpace human creation becomes likely. This trend poses a risk of exhausting the pool of human-created training data, potentially forcing generative machine learning models to increasingly depend on their random input functions for generating novel content. In contrast to a future of content generation guided by aimless random functions, our approach allows for individualized creative exploration, ensuring that computer-assisted content generation methods are human-centric and culturally relevant through time.
2201.01422
Chongjun Ouyang
Chongjun Ouyang, Yuanwei Liu, and Hongwen Yang
On the Performance of Uplink ISAC Systems
5 pages
null
null
null
cs.IT eess.SP math.IT
http://creativecommons.org/publicdomain/zero/1.0/
This letter analyzes the performance of uplink integrated sensing and communications (ISAC) systems where communication users (CUs) and radar targets (RTs) share the same frequency band. A non-orthogonal multiple access (NOMA) protocol is adopted in the communication procedure of the ISAC system. Novel expressions are derived to characterize the outage probability, ergodic communication rate, and sensing rate. Besides, the diversity order and high signal-to-noise ratio (SNR) slope are unveiled to gain further insights. It is found that when achieving the same communication rate, the ISAC system enjoys a higher sensing rate than the conventional frequency-division sensing and communications (FDSAC) system where CUs and RTs share isolated bands. All the results are validated by numerical simulations and are in excellent agreement.
[ { "created": "Wed, 5 Jan 2022 02:56:34 GMT", "version": "v1" }, { "created": "Fri, 27 May 2022 03:54:35 GMT", "version": "v2" } ]
2022-05-30
[ [ "Ouyang", "Chongjun", "" ], [ "Liu", "Yuanwei", "" ], [ "Yang", "Hongwen", "" ] ]
This letter analyzes the performance of uplink integrated sensing and communications (ISAC) systems where communication users (CUs) and radar targets (RTs) share the same frequency band. A non-orthogonal multiple access (NOMA) protocol is adopted in the communication procedure of the ISAC system. Novel expressions are derived to characterize the outage probability, ergodic communication rate, and sensing rate. Besides, the diversity order and high signal-to-noise ratio (SNR) slope are unveiled to gain further insights. It is found that when achieving the same communication rate, the ISAC system enjoys a higher sensing rate than the conventional frequency-division sensing and communications (FDSAC) system where CUs and RTs share isolated bands. All the results are validated by numerical simulations and are in excellent agreement.
2105.12524
Caglar Demir
Caglar Demir and Axel-Cyrille Ngonga Ngomo
Out-of-Vocabulary Entities in Link Prediction
null
null
null
null
cs.LG cs.SI
http://creativecommons.org/licenses/by-nc-nd/4.0/
Knowledge graph embedding techniques are key to making knowledge graphs amenable to the plethora of machine learning approaches based on vector representations. Link prediction is often used as a proxy to evaluate the quality of these embeddings. Given that the creation of benchmarks for link prediction is a time-consuming endeavor, most work on the subject matter uses only a few benchmarks. As benchmarks are crucial for the fair comparison of algorithms, ensuring their quality is tantamount to providing a solid ground for developing better solutions to link prediction and ipso facto embedding knowledge graphs. First studies of benchmarks pointed to limitations pertaining to information leaking from the development to the test fragments of some benchmark datasets. We spotted a further common limitation of three of the benchmarks commonly used for evaluating link prediction approaches: out-of-vocabulary entities in the test and validation sets. We provide an implementation of an approach for spotting and removing such entities and provide corrected versions of the datasets WN18RR, FB15K-237, and YAGO3-10. Our experiments on the corrected versions of WN18RR, FB15K-237, and YAGO3-10 suggest that the measured performance of state-of-the-art approaches is altered significantly with p-values <1%, <1.4%, and <1%, respectively. Overall, state-of-the-art approaches gain on average absolute $3.29 \pm 0.24\%$ in all metrics on WN18RR. This means that some of the conclusions achieved in previous works might need to be revisited. We provide an open-source implementation of our experiments and corrected datasets at at https://github.com/dice-group/OOV-In-Link-Prediction.
[ { "created": "Wed, 26 May 2021 12:58:18 GMT", "version": "v1" } ]
2021-05-27
[ [ "Demir", "Caglar", "" ], [ "Ngomo", "Axel-Cyrille Ngonga", "" ] ]
Knowledge graph embedding techniques are key to making knowledge graphs amenable to the plethora of machine learning approaches based on vector representations. Link prediction is often used as a proxy to evaluate the quality of these embeddings. Given that the creation of benchmarks for link prediction is a time-consuming endeavor, most work on the subject matter uses only a few benchmarks. As benchmarks are crucial for the fair comparison of algorithms, ensuring their quality is tantamount to providing a solid ground for developing better solutions to link prediction and ipso facto embedding knowledge graphs. First studies of benchmarks pointed to limitations pertaining to information leaking from the development to the test fragments of some benchmark datasets. We spotted a further common limitation of three of the benchmarks commonly used for evaluating link prediction approaches: out-of-vocabulary entities in the test and validation sets. We provide an implementation of an approach for spotting and removing such entities and provide corrected versions of the datasets WN18RR, FB15K-237, and YAGO3-10. Our experiments on the corrected versions of WN18RR, FB15K-237, and YAGO3-10 suggest that the measured performance of state-of-the-art approaches is altered significantly with p-values <1%, <1.4%, and <1%, respectively. Overall, state-of-the-art approaches gain on average absolute $3.29 \pm 0.24\%$ in all metrics on WN18RR. This means that some of the conclusions achieved in previous works might need to be revisited. We provide an open-source implementation of our experiments and corrected datasets at at https://github.com/dice-group/OOV-In-Link-Prediction.
1403.1362
Shireesha Chintalapati
Shireesha Chintalapati and M. V. Raghunadh
Illumination,Expression and Occlusion Invariant Pose-Adaptive Face Recognition System for Real-Time Applications
7 pages,8 figures, Published with International Journal of Engineering Trends and Technology (IJETT)
International Journal of Engineering Trends and Technology(IJETT), V8(6),292-298 February 2014. Published by seventh sense research group
10.14445/22315381/IJETT-V8P254
null
cs.CV
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
Face recognition in real-time scenarios is mainly affected by illumination, expression and pose variations and also by occlusion. This paper presents the framework for pose adaptive component-based face recognition system. The framework proposed deals with all the above mentioned issues. The steps involved in the presented framework are (i) facial landmark localisation, (ii) facial component extraction, (iii) pre-processing of facial image (iv) facial pose estimation (v) feature extraction using Local Binary Pattern Histograms of each component followed by (vi) fusion of pose adaptive classification of components. By employing pose adaptive classification, the recognition process is carried out on some part of database, based on estimated pose, instead of applying the recognition process on the whole database. Pre-processing techniques employed to overcome the problems due to illumination variation are also discussed in this paper. Component-based techniques provide better recognition rates when face images are occluded compared to the holistic methods. Our method is simple, feasible and provides better results when compared to other holistic methods.
[ { "created": "Thu, 6 Mar 2014 07:19:24 GMT", "version": "v1" } ]
2014-03-07
[ [ "Chintalapati", "Shireesha", "" ], [ "Raghunadh", "M. V.", "" ] ]
Face recognition in real-time scenarios is mainly affected by illumination, expression and pose variations and also by occlusion. This paper presents the framework for pose adaptive component-based face recognition system. The framework proposed deals with all the above mentioned issues. The steps involved in the presented framework are (i) facial landmark localisation, (ii) facial component extraction, (iii) pre-processing of facial image (iv) facial pose estimation (v) feature extraction using Local Binary Pattern Histograms of each component followed by (vi) fusion of pose adaptive classification of components. By employing pose adaptive classification, the recognition process is carried out on some part of database, based on estimated pose, instead of applying the recognition process on the whole database. Pre-processing techniques employed to overcome the problems due to illumination variation are also discussed in this paper. Component-based techniques provide better recognition rates when face images are occluded compared to the holistic methods. Our method is simple, feasible and provides better results when compared to other holistic methods.
2311.18488
Sana Javed
Sana Javed, Francisco Garcia-Herrero, Bane Vasic, Mark F. Flanagan
Low-Complexity Linear Programming Based Decoding of Quantum LDPC codes
Accepted for publication at the IEEE International Conference on Communications (ICC) 2024
null
null
null
cs.IT math.IT
http://creativecommons.org/licenses/by/4.0/
This paper proposes two approaches for reducing the impact of the error floor phenomenon when decoding quantum low-density parity-check codes with belief propagation based algorithms. First, a low-complexity syndrome-based linear programming (SB-LP) decoding algorithm is proposed, and second, the proposed SB-LP is applied as a post-processing step after syndrome-based min-sum (SB-MS) decoding. For the latter case, a new early stopping criterion is introduced to decide when to activate the SB-LP algorithm, avoiding executing a predefined maximum number of iterations for the SB-MS decoder. Simulation results show, for a sample hypergraph code, that the proposed decoder can lower the error floor by two to three orders of magnitude compared to SB-MS for the same total number of decoding iterations.
[ { "created": "Thu, 30 Nov 2023 12:01:04 GMT", "version": "v1" }, { "created": "Fri, 19 Jan 2024 15:53:11 GMT", "version": "v2" } ]
2024-01-22
[ [ "Javed", "Sana", "" ], [ "Garcia-Herrero", "Francisco", "" ], [ "Vasic", "Bane", "" ], [ "Flanagan", "Mark F.", "" ] ]
This paper proposes two approaches for reducing the impact of the error floor phenomenon when decoding quantum low-density parity-check codes with belief propagation based algorithms. First, a low-complexity syndrome-based linear programming (SB-LP) decoding algorithm is proposed, and second, the proposed SB-LP is applied as a post-processing step after syndrome-based min-sum (SB-MS) decoding. For the latter case, a new early stopping criterion is introduced to decide when to activate the SB-LP algorithm, avoiding executing a predefined maximum number of iterations for the SB-MS decoder. Simulation results show, for a sample hypergraph code, that the proposed decoder can lower the error floor by two to three orders of magnitude compared to SB-MS for the same total number of decoding iterations.
1511.07792
Elena Dubrova
Elena Dubrova and Mats N\"aslund and Gunnar Carlsson and John Fornehed and Ben Smeets
Two Countermeasures Against Hardware Trojans Exploiting Non-Zero Aliasing Probability of BIST
16 pages, 5 figures
null
null
null
cs.CR
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
The threat of hardware Trojans has been widely recognized by academia, industry, and government agencies. A Trojan can compromise security of a system in spite of cryptographic protection. The damage caused by a Trojan may not be limited to a business or reputation, but could have a severe impact on public safety, national economy, or national security. An extremely stealthy way of implementing hardware Trojans has been presented by Becker et al. at CHES'2012. Their work have shown that it is possible to inject a Trojan in a random number generator compliant with FIPS 140-2 and NIST SP800-90 standards by exploiting non-zero aliasing probability of Logic Built-In-Self-Test (LBIST). In this paper, we present two methods for modifying LBIST to prevent such an attack. The first method makes test patterns dependent on a configurable key which is programed into a chip after the manufacturing stage. The second method uses a remote test management system which can execute LBIST using a different set of test patterns at each test cycle.
[ { "created": "Tue, 24 Nov 2015 16:40:08 GMT", "version": "v1" } ]
2015-11-25
[ [ "Dubrova", "Elena", "" ], [ "Näslund", "Mats", "" ], [ "Carlsson", "Gunnar", "" ], [ "Fornehed", "John", "" ], [ "Smeets", "Ben", "" ] ]
The threat of hardware Trojans has been widely recognized by academia, industry, and government agencies. A Trojan can compromise security of a system in spite of cryptographic protection. The damage caused by a Trojan may not be limited to a business or reputation, but could have a severe impact on public safety, national economy, or national security. An extremely stealthy way of implementing hardware Trojans has been presented by Becker et al. at CHES'2012. Their work have shown that it is possible to inject a Trojan in a random number generator compliant with FIPS 140-2 and NIST SP800-90 standards by exploiting non-zero aliasing probability of Logic Built-In-Self-Test (LBIST). In this paper, we present two methods for modifying LBIST to prevent such an attack. The first method makes test patterns dependent on a configurable key which is programed into a chip after the manufacturing stage. The second method uses a remote test management system which can execute LBIST using a different set of test patterns at each test cycle.
2006.13833
Luis Lastras
Luis A. Lastras
Lattice Representation Learning
null
null
null
null
cs.LG cs.IT math.IT stat.ML
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
In this article we introduce theory and algorithms for learning discrete representations that take on a lattice that is embedded in an Euclidean space. Lattice representations possess an interesting combination of properties: a) they can be computed explicitly using lattice quantization, yet they can be learned efficiently using the ideas we introduce in this paper, b) they are highly related to Gaussian Variational Autoencoders, allowing designers familiar with the latter to easily produce discrete representations from their models and c) since lattices satisfy the axioms of a group, their adoption can lead into a way of learning simple algebras for modeling binary operations between objects through symbolic formalisms, yet learn these structures also formally using differentiation techniques. This article will focus on laying the groundwork for exploring and exploiting the first two properties, including a new mathematical result linking expressions used during training and inference time and experimental validation on two popular datasets.
[ { "created": "Wed, 24 Jun 2020 16:05:11 GMT", "version": "v1" } ]
2020-06-25
[ [ "Lastras", "Luis A.", "" ] ]
In this article we introduce theory and algorithms for learning discrete representations that take on a lattice that is embedded in an Euclidean space. Lattice representations possess an interesting combination of properties: a) they can be computed explicitly using lattice quantization, yet they can be learned efficiently using the ideas we introduce in this paper, b) they are highly related to Gaussian Variational Autoencoders, allowing designers familiar with the latter to easily produce discrete representations from their models and c) since lattices satisfy the axioms of a group, their adoption can lead into a way of learning simple algebras for modeling binary operations between objects through symbolic formalisms, yet learn these structures also formally using differentiation techniques. This article will focus on laying the groundwork for exploring and exploiting the first two properties, including a new mathematical result linking expressions used during training and inference time and experimental validation on two popular datasets.
2111.04671
David Pujol
David Pujol, Ashwin Machanavajjhala
Equity and Privacy: More Than Just a Tradeoff
3 pages, 1 figure. Published in IEEE Security & Privacy ( Volume: 19, Issue: 6, Nov.-Dec. 2021)
null
10.1109/MSEC.2021.3105773
null
cs.CY cs.CR
http://creativecommons.org/licenses/by/4.0/
While the entire field of privacy preserving data analytics is focused on the privacy-utility tradeoff, recent work has shown that privacy preserving data publishing can introduce different levels of utility across different population groups. It is important to understand this new tradeoff between privacy and equity as privacy technology is being deployed in situations where the data products will be used for research and policy making. Will marginal populations see disproportionately less utility from privacy technology? If there is an inequity how can we address it?
[ { "created": "Mon, 8 Nov 2021 17:39:32 GMT", "version": "v1" } ]
2021-11-09
[ [ "Pujol", "David", "" ], [ "Machanavajjhala", "Ashwin", "" ] ]
While the entire field of privacy preserving data analytics is focused on the privacy-utility tradeoff, recent work has shown that privacy preserving data publishing can introduce different levels of utility across different population groups. It is important to understand this new tradeoff between privacy and equity as privacy technology is being deployed in situations where the data products will be used for research and policy making. Will marginal populations see disproportionately less utility from privacy technology? If there is an inequity how can we address it?
2103.03443
Riccardo Paccagnella
Riccardo Paccagnella and Licheng Luo and Christopher W. Fletcher
Lord of the Ring(s): Side Channel Attacks on the CPU On-Chip Ring Interconnect Are Practical
This is the extended version of a paper that appears in USENIX Security 2021
null
null
null
cs.CR cs.AR
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
We introduce the first microarchitectural side channel attacks that leverage contention on the CPU ring interconnect. There are two challenges that make it uniquely difficult to exploit this channel. First, little is known about the ring interconnect's functioning and architecture. Second, information that can be learned by an attacker through ring contention is noisy by nature and has coarse spatial granularity. To address the first challenge, we perform a thorough reverse engineering of the sophisticated protocols that handle communication on the ring interconnect. With this knowledge, we build a cross-core covert channel over the ring interconnect with a capacity of over 4 Mbps from a single thread, the largest to date for a cross-core channel not relying on shared memory. To address the second challenge, we leverage the fine-grained temporal patterns of ring contention to infer a victim program's secrets. We demonstrate our attack by extracting key bits from vulnerable EdDSA and RSA implementations, as well as inferring the precise timing of keystrokes typed by a victim user.
[ { "created": "Fri, 5 Mar 2021 02:44:20 GMT", "version": "v1" } ]
2021-03-08
[ [ "Paccagnella", "Riccardo", "" ], [ "Luo", "Licheng", "" ], [ "Fletcher", "Christopher W.", "" ] ]
We introduce the first microarchitectural side channel attacks that leverage contention on the CPU ring interconnect. There are two challenges that make it uniquely difficult to exploit this channel. First, little is known about the ring interconnect's functioning and architecture. Second, information that can be learned by an attacker through ring contention is noisy by nature and has coarse spatial granularity. To address the first challenge, we perform a thorough reverse engineering of the sophisticated protocols that handle communication on the ring interconnect. With this knowledge, we build a cross-core covert channel over the ring interconnect with a capacity of over 4 Mbps from a single thread, the largest to date for a cross-core channel not relying on shared memory. To address the second challenge, we leverage the fine-grained temporal patterns of ring contention to infer a victim program's secrets. We demonstrate our attack by extracting key bits from vulnerable EdDSA and RSA implementations, as well as inferring the precise timing of keystrokes typed by a victim user.
1609.09267
Andrea Romanoni
Gheorghii Postica and Andrea Romanoni and Matteo Matteucci
Robust Moving Objects Detection in Lidar Data Exploiting Visual Cues
6 pages, to appear in IROS 2016
null
null
null
cs.RO cs.CV
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
Detecting moving objects in dynamic scenes from sequences of lidar scans is an important task in object tracking, mapping, localization, and navigation. Many works focus on changes detection in previously observed scenes, while a very limited amount of literature addresses moving objects detection. The state-of-the-art method exploits Dempster-Shafer Theory to evaluate the occupancy of a lidar scan and to discriminate points belonging to the static scene from moving ones. In this paper we improve both speed and accuracy of this method by discretizing the occupancy representation, and by removing false positives through visual cues. Many false positives lying on the ground plane are also removed thanks to a novel ground plane removal algorithm. Efficiency is improved through an octree indexing strategy. Experimental evaluation against the KITTI public dataset shows the effectiveness of our approach, both qualitatively and quantitatively with respect to the state- of-the-art.
[ { "created": "Thu, 29 Sep 2016 09:29:46 GMT", "version": "v1" } ]
2016-09-30
[ [ "Postica", "Gheorghii", "" ], [ "Romanoni", "Andrea", "" ], [ "Matteucci", "Matteo", "" ] ]
Detecting moving objects in dynamic scenes from sequences of lidar scans is an important task in object tracking, mapping, localization, and navigation. Many works focus on changes detection in previously observed scenes, while a very limited amount of literature addresses moving objects detection. The state-of-the-art method exploits Dempster-Shafer Theory to evaluate the occupancy of a lidar scan and to discriminate points belonging to the static scene from moving ones. In this paper we improve both speed and accuracy of this method by discretizing the occupancy representation, and by removing false positives through visual cues. Many false positives lying on the ground plane are also removed thanks to a novel ground plane removal algorithm. Efficiency is improved through an octree indexing strategy. Experimental evaluation against the KITTI public dataset shows the effectiveness of our approach, both qualitatively and quantitatively with respect to the state- of-the-art.
2210.17087
Youpeng Zhao
Yudong Lu, Jian Zhao, Youpeng Zhao, Wengang Zhou, Houqiang Li
DanZero: Mastering GuanDan Game with Reinforcement Learning
null
null
null
null
cs.AI cs.LG
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
Card game AI has always been a hot topic in the research of artificial intelligence. In recent years, complex card games such as Mahjong, DouDizhu and Texas Hold'em have been solved and the corresponding AI programs have reached the level of human experts. In this paper, we are devoted to developing an AI program for a more complex card game, GuanDan, whose rules are similar to DouDizhu but much more complicated. To be specific, the characteristics of large state and action space, long length of one episode and the unsure number of players in the GuanDan pose great challenges for the development of the AI program. To address these issues, we propose the first AI program DanZero for GuanDan using reinforcement learning technique. Specifically, we utilize a distributed framework to train our AI system. In the actor processes, we carefully design the state features and agents generate samples by self-play. In the learner process, the model is updated by Deep Monte-Carlo Method. After training for 30 days using 160 CPUs and 1 GPU, we get our DanZero bot. We compare it with 8 baseline AI programs which are based on heuristic rules and the results reveal the outstanding performance of DanZero. We also test DanZero with human players and demonstrate its human-level performance.
[ { "created": "Mon, 31 Oct 2022 06:29:08 GMT", "version": "v1" } ]
2022-11-01
[ [ "Lu", "Yudong", "" ], [ "Zhao", "Jian", "" ], [ "Zhao", "Youpeng", "" ], [ "Zhou", "Wengang", "" ], [ "Li", "Houqiang", "" ] ]
Card game AI has always been a hot topic in the research of artificial intelligence. In recent years, complex card games such as Mahjong, DouDizhu and Texas Hold'em have been solved and the corresponding AI programs have reached the level of human experts. In this paper, we are devoted to developing an AI program for a more complex card game, GuanDan, whose rules are similar to DouDizhu but much more complicated. To be specific, the characteristics of large state and action space, long length of one episode and the unsure number of players in the GuanDan pose great challenges for the development of the AI program. To address these issues, we propose the first AI program DanZero for GuanDan using reinforcement learning technique. Specifically, we utilize a distributed framework to train our AI system. In the actor processes, we carefully design the state features and agents generate samples by self-play. In the learner process, the model is updated by Deep Monte-Carlo Method. After training for 30 days using 160 CPUs and 1 GPU, we get our DanZero bot. We compare it with 8 baseline AI programs which are based on heuristic rules and the results reveal the outstanding performance of DanZero. We also test DanZero with human players and demonstrate its human-level performance.
2406.12896
Wei Zhang
Jiajun Cui, Hong Qian, Bo Jiang, Wei Zhang
Leveraging Pedagogical Theories to Understand Student Learning Process with Graph-based Reasonable Knowledge Tracing
Preprint, accepted to appear in SIGKDD 2024, 12 pages. The source code is available at https://github.com/JJCui96/GRKT. Keywords: interpretable knowledge tracing, student behavior modeling, intelligence education
null
null
null
cs.AI cs.CY cs.LG
http://creativecommons.org/licenses/by-nc-sa/4.0/
Knowledge tracing (KT) is a crucial task in intelligent education, focusing on predicting students' performance on given questions to trace their evolving knowledge. The advancement of deep learning in this field has led to deep-learning knowledge tracing (DLKT) models that prioritize high predictive accuracy. However, many existing DLKT methods overlook the fundamental goal of tracking students' dynamical knowledge mastery. These models do not explicitly model knowledge mastery tracing processes or yield unreasonable results that educators find difficulty to comprehend and apply in real teaching scenarios. In response, our research conducts a preliminary analysis of mainstream KT approaches to highlight and explain such unreasonableness. We introduce GRKT, a graph-based reasonable knowledge tracing method to address these issues. By leveraging graph neural networks, our approach delves into the mutual influences of knowledge concepts, offering a more accurate representation of how the knowledge mastery evolves throughout the learning process. Additionally, we propose a fine-grained and psychological three-stage modeling process as knowledge retrieval, memory strengthening, and knowledge learning/forgetting, to conduct a more reasonable knowledge tracing process. Comprehensive experiments demonstrate that GRKT outperforms eleven baselines across three datasets, not only enhancing predictive accuracy but also generating more reasonable knowledge tracing results. This makes our model a promising advancement for practical implementation in educational settings. The source code is available at https://github.com/JJCui96/GRKT.
[ { "created": "Fri, 7 Jun 2024 10:14:30 GMT", "version": "v1" } ]
2024-06-21
[ [ "Cui", "Jiajun", "" ], [ "Qian", "Hong", "" ], [ "Jiang", "Bo", "" ], [ "Zhang", "Wei", "" ] ]
Knowledge tracing (KT) is a crucial task in intelligent education, focusing on predicting students' performance on given questions to trace their evolving knowledge. The advancement of deep learning in this field has led to deep-learning knowledge tracing (DLKT) models that prioritize high predictive accuracy. However, many existing DLKT methods overlook the fundamental goal of tracking students' dynamical knowledge mastery. These models do not explicitly model knowledge mastery tracing processes or yield unreasonable results that educators find difficulty to comprehend and apply in real teaching scenarios. In response, our research conducts a preliminary analysis of mainstream KT approaches to highlight and explain such unreasonableness. We introduce GRKT, a graph-based reasonable knowledge tracing method to address these issues. By leveraging graph neural networks, our approach delves into the mutual influences of knowledge concepts, offering a more accurate representation of how the knowledge mastery evolves throughout the learning process. Additionally, we propose a fine-grained and psychological three-stage modeling process as knowledge retrieval, memory strengthening, and knowledge learning/forgetting, to conduct a more reasonable knowledge tracing process. Comprehensive experiments demonstrate that GRKT outperforms eleven baselines across three datasets, not only enhancing predictive accuracy but also generating more reasonable knowledge tracing results. This makes our model a promising advancement for practical implementation in educational settings. The source code is available at https://github.com/JJCui96/GRKT.
1507.03851
Enric Rodriguez Carbonell
Marc Brockschmidt, Daniel Larraz, Albert Oliveras, Enric Rodriguez-Carbonell, Albert Rubio
Compositional Safety Verification with Max-SMT
Extended technical report version of the conference paper at FMCAD'15
null
null
null
cs.LO
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
We present an automated compositional program verification technique for safety properties based on conditional inductive invariants. For a given program part (e.g., a single loop) and a postcondition $\varphi$, we show how to, using a Max-SMT solver, an inductive invariant together with a precondition can be synthesized so that the precondition ensures the validity of the invariant and that the invariant implies $\varphi$. From this, we build a bottom-up program verification framework that propagates preconditions of small program parts as postconditions for preceding program parts. The method recovers from failures to prove the validity of a precondition, using the obtained intermediate results to restrict the search space for further proof attempts. As only small program parts need to be handled at a time, our method is scalable and distributable. The derived conditions can be viewed as implicit contracts between different parts of the program, and thus enable an incremental program analysis.
[ { "created": "Tue, 14 Jul 2015 14:01:56 GMT", "version": "v1" }, { "created": "Wed, 15 Jul 2015 05:41:20 GMT", "version": "v2" }, { "created": "Mon, 3 Aug 2015 21:41:23 GMT", "version": "v3" } ]
2015-08-05
[ [ "Brockschmidt", "Marc", "" ], [ "Larraz", "Daniel", "" ], [ "Oliveras", "Albert", "" ], [ "Rodriguez-Carbonell", "Enric", "" ], [ "Rubio", "Albert", "" ] ]
We present an automated compositional program verification technique for safety properties based on conditional inductive invariants. For a given program part (e.g., a single loop) and a postcondition $\varphi$, we show how to, using a Max-SMT solver, an inductive invariant together with a precondition can be synthesized so that the precondition ensures the validity of the invariant and that the invariant implies $\varphi$. From this, we build a bottom-up program verification framework that propagates preconditions of small program parts as postconditions for preceding program parts. The method recovers from failures to prove the validity of a precondition, using the obtained intermediate results to restrict the search space for further proof attempts. As only small program parts need to be handled at a time, our method is scalable and distributable. The derived conditions can be viewed as implicit contracts between different parts of the program, and thus enable an incremental program analysis.
1910.09787
Congcong Miao
Congcong Miao and Jilong Wang and Shuying Zhuang and Changqing An
A Coordinated View of Cyberspace
null
null
null
null
cs.NI
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
Cyberspace is an online world created by growing network of computing and communication technologies. It is a virtual space of the Internet, paralleled to geographic space we are living on. As becoming a recognized component of our society, cyberspace gradually draws more attention in academic research. Many prior efforts have tried to represent and visualize cyberspace in geographic coordinate system (GCS) and network coordinate system (NCS). However, there are some disadvantages on these views. Firstly, mapping cyberspace in geographic space only reveals a partial characteristics of cyberspace, especially geographic characteristic of cyberspace. All what we could see is only the geographic information of cyberspace and tip of the iceberg of cyberspace. Secondly, NCS is established according to network topology and maps the position of each node in the coordinate system according to RTT (Round Trip Time) or network delays. However, this coordinate system changes dynamically with RTT changes or host connection status, resulting in the coordinate system not stable. Cyberspace, regarded as a second space in human life, is rather complex and multi-dimension. However, it is little known to us. It is in a great need of establishing its own coordinate system to tackle the challenging task of efficiently visualizing complex multi-dimensional cyberspace and get to know more about cyberspace. This paper aims to explore and visualize cyberspace. To best of our knowledge, we are firstly to establish a Cyberspace Coordination System (CyberCS) to represent and visualize cyberspace. CyberCS will make the representation of cyberspace easier or more concrete which is similar to Fourier transform. With the help of CyberCS, different parts and degrees of cyberspace are efficiently visualized and user can easily filter out the specific details of interest.
[ { "created": "Tue, 22 Oct 2019 06:50:02 GMT", "version": "v1" } ]
2019-10-23
[ [ "Miao", "Congcong", "" ], [ "Wang", "Jilong", "" ], [ "Zhuang", "Shuying", "" ], [ "An", "Changqing", "" ] ]
Cyberspace is an online world created by growing network of computing and communication technologies. It is a virtual space of the Internet, paralleled to geographic space we are living on. As becoming a recognized component of our society, cyberspace gradually draws more attention in academic research. Many prior efforts have tried to represent and visualize cyberspace in geographic coordinate system (GCS) and network coordinate system (NCS). However, there are some disadvantages on these views. Firstly, mapping cyberspace in geographic space only reveals a partial characteristics of cyberspace, especially geographic characteristic of cyberspace. All what we could see is only the geographic information of cyberspace and tip of the iceberg of cyberspace. Secondly, NCS is established according to network topology and maps the position of each node in the coordinate system according to RTT (Round Trip Time) or network delays. However, this coordinate system changes dynamically with RTT changes or host connection status, resulting in the coordinate system not stable. Cyberspace, regarded as a second space in human life, is rather complex and multi-dimension. However, it is little known to us. It is in a great need of establishing its own coordinate system to tackle the challenging task of efficiently visualizing complex multi-dimensional cyberspace and get to know more about cyberspace. This paper aims to explore and visualize cyberspace. To best of our knowledge, we are firstly to establish a Cyberspace Coordination System (CyberCS) to represent and visualize cyberspace. CyberCS will make the representation of cyberspace easier or more concrete which is similar to Fourier transform. With the help of CyberCS, different parts and degrees of cyberspace are efficiently visualized and user can easily filter out the specific details of interest.
2210.14505
Xiujing Zheng
Xiujing Zheng and Liqi Wang and Shixin Zhu
Constructions of entanglement-assisted quantum MDS codes from generalized Reed-Solomon codes
21 pages. 5 table
null
null
null
cs.IT math.IT
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
By generalizing the stabilizer quantum error-correcting codes, entanglement-assisted quantum error-correcting (EAQEC) codes were introduced, which could be derived from any classical linear codes via the relaxation of self-orthogonality conditions with the aid of pre-shared entanglement between the sender and the receiver. In this paper, three classes of entanglement-assisted quantum error-correcting maximum-distance-separable (EAQMDS) codes are constructed through generalized Reed-Solomon codes. Under our constructions, the minimum distances of our EAQMDS codes are much larger than those of the known EAQMDS codes of the same lengths that consume the same number of ebits. Furthermore, some of the lengths of the EAQMDS codes are not divisors of $q^2-1$, which are completely new and unlike all those known lengths existed before.
[ { "created": "Wed, 26 Oct 2022 06:30:15 GMT", "version": "v1" }, { "created": "Sat, 16 Mar 2024 07:26:52 GMT", "version": "v2" } ]
2024-03-19
[ [ "Zheng", "Xiujing", "" ], [ "Wang", "Liqi", "" ], [ "Zhu", "Shixin", "" ] ]
By generalizing the stabilizer quantum error-correcting codes, entanglement-assisted quantum error-correcting (EAQEC) codes were introduced, which could be derived from any classical linear codes via the relaxation of self-orthogonality conditions with the aid of pre-shared entanglement between the sender and the receiver. In this paper, three classes of entanglement-assisted quantum error-correcting maximum-distance-separable (EAQMDS) codes are constructed through generalized Reed-Solomon codes. Under our constructions, the minimum distances of our EAQMDS codes are much larger than those of the known EAQMDS codes of the same lengths that consume the same number of ebits. Furthermore, some of the lengths of the EAQMDS codes are not divisors of $q^2-1$, which are completely new and unlike all those known lengths existed before.
1610.04591
Peter LeFanu Lumsdaine
Andrej Bauer, Jason Gross, Peter LeFanu Lumsdaine, Mike Shulman, Matthieu Sozeau, and Bas Spitters
The HoTT Library: A formalization of homotopy type theory in Coq
null
null
null
null
cs.LO math.LO
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
We report on the development of the HoTT library, a formalization of homotopy type theory in the Coq proof assistant. It formalizes most of basic homotopy type theory, including univalence, higher inductive types, and significant amounts of synthetic homotopy theory, as well as category theory and modalities. The library has been used as a basis for several independent developments. We discuss the decisions that led to the design of the library, and we comment on the interaction of homotopy type theory with recently introduced features of Coq, such as universe polymorphism and private inductive types.
[ { "created": "Fri, 14 Oct 2016 19:23:50 GMT", "version": "v1" }, { "created": "Fri, 9 Dec 2016 16:31:04 GMT", "version": "v2" } ]
2017-05-02
[ [ "Bauer", "Andrej", "" ], [ "Gross", "Jason", "" ], [ "Lumsdaine", "Peter LeFanu", "" ], [ "Shulman", "Mike", "" ], [ "Sozeau", "Matthieu", "" ], [ "Spitters", "Bas", "" ] ]
We report on the development of the HoTT library, a formalization of homotopy type theory in the Coq proof assistant. It formalizes most of basic homotopy type theory, including univalence, higher inductive types, and significant amounts of synthetic homotopy theory, as well as category theory and modalities. The library has been used as a basis for several independent developments. We discuss the decisions that led to the design of the library, and we comment on the interaction of homotopy type theory with recently introduced features of Coq, such as universe polymorphism and private inductive types.
1805.11462
Vincent Nguyen
Guillaume Klein, Yoon Kim, Yuntian Deng, Vincent Nguyen, Jean Senellart, Alexander M. Rush
OpenNMT: Neural Machine Translation Toolkit
Presentation to AMTA 2018 - Boston. arXiv admin note: substantial text overlap with arXiv:1701.02810
null
null
null
cs.CL
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
OpenNMT is an open-source toolkit for neural machine translation (NMT). The system prioritizes efficiency, modularity, and extensibility with the goal of supporting NMT research into model architectures, feature representations, and source modalities, while maintaining competitive performance and reasonable training requirements. The toolkit consists of modeling and translation support, as well as detailed pedagogical documentation about the underlying techniques. OpenNMT has been used in several production MT systems, modified for numerous research papers, and is implemented across several deep learning frameworks.
[ { "created": "Mon, 28 May 2018 07:58:46 GMT", "version": "v1" } ]
2018-05-30
[ [ "Klein", "Guillaume", "" ], [ "Kim", "Yoon", "" ], [ "Deng", "Yuntian", "" ], [ "Nguyen", "Vincent", "" ], [ "Senellart", "Jean", "" ], [ "Rush", "Alexander M.", "" ] ]
OpenNMT is an open-source toolkit for neural machine translation (NMT). The system prioritizes efficiency, modularity, and extensibility with the goal of supporting NMT research into model architectures, feature representations, and source modalities, while maintaining competitive performance and reasonable training requirements. The toolkit consists of modeling and translation support, as well as detailed pedagogical documentation about the underlying techniques. OpenNMT has been used in several production MT systems, modified for numerous research papers, and is implemented across several deep learning frameworks.
1702.07492
Ahmed Qureshi
Ahmed Hussain Qureshi, Yutaka Nakamura, Yuichiro Yoshikawa and Hiroshi Ishiguro
Robot gains Social Intelligence through Multimodal Deep Reinforcement Learning
The paper is published in IEEE-RAS International Conference on Humanoid Robots (Humanoids) 2016
null
null
null
cs.RO cs.AI cs.CV stat.ML
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
For robots to coexist with humans in a social world like ours, it is crucial that they possess human-like social interaction skills. Programming a robot to possess such skills is a challenging task. In this paper, we propose a Multimodal Deep Q-Network (MDQN) to enable a robot to learn human-like interaction skills through a trial and error method. This paper aims to develop a robot that gathers data during its interaction with a human and learns human interaction behaviour from the high-dimensional sensory information using end-to-end reinforcement learning. This paper demonstrates that the robot was able to learn basic interaction skills successfully, after 14 days of interacting with people.
[ { "created": "Fri, 24 Feb 2017 08:30:43 GMT", "version": "v1" } ]
2017-02-27
[ [ "Qureshi", "Ahmed Hussain", "" ], [ "Nakamura", "Yutaka", "" ], [ "Yoshikawa", "Yuichiro", "" ], [ "Ishiguro", "Hiroshi", "" ] ]
For robots to coexist with humans in a social world like ours, it is crucial that they possess human-like social interaction skills. Programming a robot to possess such skills is a challenging task. In this paper, we propose a Multimodal Deep Q-Network (MDQN) to enable a robot to learn human-like interaction skills through a trial and error method. This paper aims to develop a robot that gathers data during its interaction with a human and learns human interaction behaviour from the high-dimensional sensory information using end-to-end reinforcement learning. This paper demonstrates that the robot was able to learn basic interaction skills successfully, after 14 days of interacting with people.