Filters








12,925 Hits in 5.4 sec

Understanding and Mitigating the Tradeoff Between Robustness and Accuracy [article]

Aditi Raghunathan, Sang Michael Xie, Fanny Yang, John Duchi, Percy Liang
2020 arXiv   pre-print
Previous explanations for this tradeoff rely on the assumption that no predictor in the hypothesis class has low standard and robust error.  ...  In this work, we precisely characterize the effect of augmentation on the standard error in linear regression when the optimal linear predictor has zero standard and robust error.  ...  All models are trained for 200 epochs with respect to the size of the labeled training dataset and all achieve almost 100% standard and robust training accuracy.  ... 
arXiv:2002.10716v2 fatcat:mq6fvnellfegzax3sohbeeitpu

Adversarial Training Can Hurt Generalization [article]

Aditi Raghunathan, Sang Michael Xie, Fanny Yang, John C. Duchi, Percy Liang
2019 arXiv   pre-print
Previous work has studied this tradeoff between standard and robust accuracy, but only in the setting where no predictor performs well on both objectives in the infinite data limit.  ...  Furthermore, since our construction is based on a convex learning problem, we rule out optimization concerns, thus laying bare a fundamental tension between robustness and generalization.  ...  FY was supported by the Institute for Theoretical Studies ETH Zurich and the Dr. Max Rössler and the Walter Haefner Foundation.  ... 
arXiv:1906.06032v2 fatcat:awfsnfzzmfbolp5zfpbqzqw26a

Indicators of Attack Failure: Debugging and Improving Optimization of Adversarial Examples [article]

Maura Pintor, Luca Demetrio, Angelo Sotgiu, Giovanni Manca, Ambra Demontis, Nicholas Carlini, Battista Biggio, Fabio Roli
2021 arXiv   pre-print
In this work, we overcome these limitations by (i) defining a set of quantitative indicators which unveil common failures in the optimization of gradient-based attacks, and (ii) proposing specific mitigation  ...  Although guidelines and best practices have been suggested to improve current adversarial robustness evaluations, the lack of automatic testing and debugging tools makes it difficult to apply these recommendations  ...  Acknowledgments and Disclosure of Funding  ... 
arXiv:2106.09947v1 fatcat:6zgz6pd2rvckbdzvmqvg2ghjlm

Measuring Adversarial Robustness using a Voronoi-Epsilon Adversary [article]

Hyeongji Kim, Pekka Parviainen, Ketil Malde
2021 arXiv   pre-print
Previous studies on robustness have argued that there is a tradeoff between accuracy and adversarial accuracy. The tradeoff can be inevitable even when we neglect generalization.  ...  As a result, adversarial accuracy based on this adversary avoids a tradeoff between accuracy and adversarial accuracy on training data even when ϵ is large.  ...  Yi Liu, and Jungeum Kim for the helpful feedback. We also thank Dr. Wieland Brendel for the helpful discussions.  ... 
arXiv:2005.02540v3 fatcat:x4pjxrf4mffdtezlxibf3uc2ra

A Philosophy for Developing Trust in Self-driving Cars [chapter]

Michael Wagner, Philip Koopman
2015 Lecture Notes in Mobility  
For decades, our lives have depended on the safe operation of automated mechanisms around and inside us. The autonomy and complexity of these mechanisms is increasing dramatically.  ...  In this paper we survey existing methods and tools that, taken together, can enable a new and more productive philosophy for software safety that is based on Karl Popper's idea of falsificationism.  ...  We are obligated to understand and to mitigate those risks that remain to the extent possible. How can we do this?  ... 
doi:10.1007/978-3-319-19078-5_14 fatcat:fewiud4pefeaxeq6cdpshivwba

FR-Train: A Mutual Information-Based Approach to Fair and Robust Training [article]

Yuji Roh, Kangwook Lee, Steven Euijong Whang, Changho Suh
2020 arXiv   pre-print
In our experiments, FR-Train shows almost no decrease in fairness and accuracy in the presence of data poisoning by both mitigating the bias and defending against poisoning.  ...  Trustworthy AI is a critical issue in machine learning where, in addition to training a model that is accurate, one must consider both fair and robust training in the presence of data bias and poisoning  ...  Unfairness mitigation usually involves some tradeoff between the model's accuracy and fairness. Most recently, generative adversarial networks (GANs) are being adapted to a fairness setting .  ... 
arXiv:2002.10234v2 fatcat:5lqxsxfkhrcivfl5imgqfckjee

Internal Feedback in Biological Control: Architectures and Examples [article]

Anish A. Sarma, Jing Shuang Li, Josefin Stenberg, Gwyneth Card, Elizabeth S. Heckscher, Narayanan Kasthuri, Terrence Sejnowski, John C. Doyle
2021 arXiv   pre-print
In this paper, we describe these very different motivating examples and introduce the concepts necessary to explain their complex IFPs, particularly the severe speed-accuracy tradeoffs that constrain the  ...  We also sketch some minimal theory for extremely simplified toy models that nevertheless highlight the importance of diversity-enabled sweet spots (DESS) in mitigating the impact of hardware tradeoffs.  ...  At a given level of analysis (multicellular on the left, molecular on on the right), individual components face tradeoffs between crucial system-level goals like speed, accuracy, efficiency, and robustness  ... 
arXiv:2110.05029v3 fatcat:lr4rhfe5frhdderadnrepkhgoi

MAT: A Multi-strength Adversarial Training Method to Mitigate Adversarial Attacks [article]

Chang Song, Hsin-Pai Cheng, Huanrui Yang, Sicheng Li, Chunpeng Wu, Qing Wu, Hai Li, Yiran Chen
2018 arXiv   pre-print
Two training structures - mixed MAT and parallel MAT - are developed to facilitate the tradeoffs between training time and memory occupation.  ...  Our results show that MAT can substantially minimize the accuracy degradation of deep learning systems to adversarial attacks on MNIST, CIFAR-10, CIFAR-100, and SVHN.  ...  platform and discuss the tradeoffs between training time, robustness, and hardware cost.  ... 
arXiv:1705.09764v2 fatcat:bsciaowcbnbphlrl7hzmo5ycpa

Data Cleaning for Accurate, Fair, and Robust Models: A Big Data - AI Integration Approach [article]

Ki Hyun Tae, Yuji Roh, Young Hun Oh, Hyunsu Kim, Steven Euijong Whang
2019 arXiv   pre-print
As machine learning is used in sensitive applications, it becomes imperative that the trained model is accurate, fair, and robust to attacks.  ...  Although a significant amount of research has been done by each community, ultimately the same datasets must be preprocessed, and there is little understanding how the techniques relate to each other and  ...  These techniques typically tradeoff some model accuracy in order to improve model fairness.  ... 
arXiv:1904.10761v1 fatcat:fcvusaqesndyxlnnndleu4pkh4

Advancing agricultural greenhouse gas quantification*

Lydia Olander, Eva Wollenberg, Francesco Tubiello, Martin Herold
2013 Environmental Research Letters  
We dedicate this special issue to the memory of Daniel Martino, a generous leader in greenhouse gas quantification and accounting from agriculture, land-use change, and forestry.  ...  and methods that will help mitigation policy and programs move forward around the world.  ...  issue and associated activities and papers, given their common desire to improve our understanding of the state of agricultural greenhouse gas (GHG) quantification and to advance ideas for building data  ... 
doi:10.1088/1748-9326/8/1/011002 fatcat:wgfikih6izgoray6ozjv6ogpai

The Interplay between Accounting and Reporting on Mitigation Contributions under the Paris Agreement

K. Levin
2018 Carbon and Climate Law Review  
This paper explores the linkages between accounting for and reporting on mitigation contributions under the Paris Agreement.  ...  Specifically, it explores the relationship between the provisions related to communicating nationally determined contributions (NDCs) under Article 4, paragraph 8; accounting for NDCs under Article 4,  ...  Interplay between Accounting and Reporting of NDCs Ideally, the guidance related to CTU, accounting, and reporting of NDCs will be robust and provide sufficient detail and accuracy so that the Paris Agreement's  ... 
doi:10.21552/cclr/2018/3/6 fatcat:vfiue2wbobe4pibx2ywh7r454u

Precise Tradeoffs in Adversarial Training for Linear Regression [article]

Adel Javanmard, Mahdi Soltanolkotabi, Hamed Hassani
2020 arXiv   pre-print
Furthermore, we precisely characterize the standard/robust accuracy and the corresponding tradeoff achieved by a contemporary mini-max adversarial training approach in a high-dimensional regime where the  ...  In particular, we characterize the fundamental tradeoff between the accuracies achievable by any algorithm regardless of computational power or size of the training data.  ...  Therefore, it is crucial to understand the tradeoff between robust and standard accuracy with adversarial training.  ... 
arXiv:2002.10477v1 fatcat:bay7cudixbgrddzdt53bgv44xq

Neural integrators for decision making: a favorable tradeoff between robustness and sensitivity

Nicholas Cain, Andrea K. Barreiro, Michael Shadlen, Eric Shea-Brown
2013 Journal of Neurophysiology  
Neural integrators for decision making: a favorable tradeoff between robustness and sensitivity.  ...  The degree of this limiting embodies a tradeoff between sensitivity to the input stream and robustness against parameter mistuning.  ...  ACKNOWLEDGMENTS We thank the reviewers for constructive comments and insights that improved this manuscript and especially for pointing out the link between the robustness operation and thresholding sets  ... 
doi:10.1152/jn.00976.2012 pmid:23446688 pmcid:PMC3653050 fatcat:r56mytket5gn5noa2kiz2lu4l4

How fair can we go in machine learning? Assessing the boundaries of fairness in decision trees [article]

Ana Valdivia, Javier Sánchez-Monedero, Jorge Casillas
2020 arXiv   pre-print
Yet, many of these fairness-aware approaches aim to obtain a unique solution to the problem, which leads to a poor understanding of the statistical limits of bias mitigation interventions.  ...  We present the first methodology that allows to explore those limits within a multi-objective framework that seeks to optimize any measure of accuracy and fairness and provides a Pareto front with the  ...  The tradeoff between accuracy and fairness due to disparate mistreatment is expressed as a threshold parameter established by the user.  ... 
arXiv:2006.12399v1 fatcat:iul4mi7bmjd7bik4df7pxekixy

Codon-by-Codon Modulation of Translational Speed and Accuracy Via mRNA Folding

Jian-Rong Yang, Xiaoshu Chen, Jianzhi Zhang, Harmit S. Malik
2014 PLoS Biology  
However, speedy elongation undermines translational accuracy because of a mechanistic tradeoff.  ...  The exquisite codon-by-codon translational modulation uncovered here is a testament to the power of natural selection in mitigating efficiency-accuracy conflicts, which are prevalent in biology.  ...  We thank the Ken Cadigan lab for assistance in the luciferase assay, Eugene Koonin for stimulating discussion, and Calum Maclean and Wenfeng Qian for valuable comments.  ... 
doi:10.1371/journal.pbio.1001910 pmid:25051069 pmcid:PMC4106722 fatcat:onndhlz6mrfzdj5b3wdn3iwipi
« Previous Showing results 1 — 15 out of 12,925 results