800 Hits in 4.6 sec

For self-supervised learning, Rationality implies generalization, provably [article]

Yamini Bansal, Gal Kaplun, Boaz Barak
2020 arXiv   pre-print
We prove a new upper bound on the generalization gap of classifiers that are obtained by first using self-supervision to learn a representation r of the training data, and then fitting a simple (e.g.,  ...  We show that our bound is non-vacuous for many popular representation-learning based classifiers on CIFAR-10 and ImageNet, including SimCLR, AMDIM and MoCo.  ...  We also thank Oracle and Microsoft for grants used for computational resources. Y.B is partially supported by MIT-IBM Watson AI Lab.  ... 
arXiv:2010.08508v1 fatcat:zhgfvnhbunfffokrrdm67ledzu

Provable Guarantees for Self-Supervised Deep Learning with Spectral Contrastive Loss [article]

Jeff Z. HaoChen, Colin Wei, Adrien Gaidon, Tengyu Ma
2022 arXiv   pre-print
Recent works in self-supervised learning have advanced the state-of-the-art by relying on the contrastive learning paradigm, which learns representations by pushing positive pairs, or similar examples  ...  In all, this work provides the first provable analysis for contrastive learning where guarantees for linear probe evaluation can apply to realistic empirical settings.  ...  Lee, Michael Xie, and Guodong Zhang for helpful discussions. CW acknowledges support from an NSF Graduate Research Fellowship. TM acknowledges support of Google Faculty Award and NSF IIS 2045685.  ... 
arXiv:2106.04156v6 fatcat:ky5nyxtdrjg5vj7he6m4wfqoea

New Millennium AI and the Convergence of History: Update of 2012 [chapter]

Jürgen Schmidhuber
2012 The Frontiers Collection  
There also has been rapid progress in not quite universal but still rather general and practical artificial recurrent neural networks for learning sequenceprocessing programs, now yielding state-of-the-art  ...  the previously largely heuristic field of General AI and embedded agents.  ...  Optimal Self-Referential General Problem Solver The recent Gödel machines [68, 72, 74] represent the first class of mathematically rigorous, general, fully self-referential, self-improving, optimally  ... 
doi:10.1007/978-3-642-32560-1_4 fatcat:rjsvw2jsk5apbg4cxnxfkzwjya

Rationalizing Text Matching: Learning Sparse Alignments via Optimal Transport [article]

Kyle Swanson, Lili Yu, Tao Lei
2020 arXiv   pre-print
Selecting input features of top relevance has become a popular method for building self-explaining models.  ...  In this work, we extend this selective rationalization approach to text matching, where the goal is to jointly select and align text pieces, such as tokens or sentences, as a justification for the downstream  ...  Acknowledgments We thank Jesse Michel, Derek Chen, Yi Yang, and the anonymous reviewers for their valuable discussions.  ... 
arXiv:2005.13111v1 fatcat:px3x7mhxmvejrbtzqmefews5f4

Franco Montagna's Work on Provability Logic and Many-valued Logic

Lev Beklemishev, Tommaso Flaminio
2016 Studia Logica: An International Journal for Symbolic Logic  
We survey some of his results and ideas in the two disciplines he greatly contributed along his career: provability logic and many-valued logic.  ...  We would like to thank the anonymous referees for careful reading and precious suggestions.  ...  The work of the first author was supported by Russian Foundation for Basic Research, project No. 15-01-09218a.  ... 
doi:10.1007/s11225-016-9654-3 fatcat:7dhk6qh25fdvfg2q3mydf4qs6m

Multi-Agent Reinforcement Learning: A Selective Overview of Theories and Algorithms [article]

Kaiqing Zhang, Zhuoran Yang, Tamer Başar
2021 arXiv   pre-print
the mean-field regime, (non-)convergence of policy-based methods for learning in games, etc.  ...  Recent years have witnessed significant advances in reinforcement learning (RL), which has registered great success in solving various sequential decision-making problems in machine learning.  ...  , with the policy mixture being learned through supervised learning.  ... 
arXiv:1911.10635v2 fatcat:ihlhtjlhnrdizbkcfzsnz5urfq

Should Robots be Obedient?

Smitha Milli, Dylan Hadfield-Menell, Anca Dragan, Stuart Russell
2017 Proceedings of the Twenty-Sixth International Joint Conference on Artificial Intelligence  
Thus, there is a tradeoff between the obedience of a robot and the value it can attain for its owner.  ...  Intuitively, obedience -- following the order that a human gives -- seems like a good property for a robot to have.  ...  Acknowledgements We thank Daniel Filan for feedback on an early draft.  ... 
doi:10.24963/ijcai.2017/662 dblp:conf/ijcai/MilliHDR17 fatcat:zhvquksfp5cf5fdvxr2qi7xhtu

Should Robots be Obedient? [article]

Smitha Milli, Dylan Hadfield-Menell, Anca Dragan, Stuart Russell
2017 arXiv   pre-print
Thus, there is a tradeoff between the obedience of a robot and the value it can attain for its owner.  ...  Intuitively, obedience -- following the order that a human gives -- seems like a good property for a robot to have.  ...  We wish to analyze R's incentive to obey H given that We first contribute a general model for this type of interaction, which we call a supervision POMDP.  ... 
arXiv:1705.09990v1 fatcat:shak3nj47vh7tginijzkzn56ha

Taking Uncertainty Seriously: Adaptive Governance and International Trade: A Rejoinder to Monica Garcia-Salmones

A. Lang, R. Cooney
2009 European journal of international law  
We are grateful, therefore, to Mónica García-Salmones for her response to our article, and are pleased to have this opportunity to clarify some aspects of our thinking and our approach that may not have  ...  While we view the WTO as a policy arena that can be used for enhancing learning, the possibility of such learning does not in our view imply unproblematic harmony between actors ' interests.  ...  , and policy choices are justifi ed as products of enlightened rationality.  ... 
doi:10.1093/ejil/chp007 fatcat:e4nx6h34ibh7zemw3tmdtvnoiy

Impossibility Results in AI: A Survey [article]

Mario Brcic, Roman V. Yampolskiy
2022 arXiv   pre-print
The remaining results deal with misalignment between the clones and put a limit to the self-awareness of agents. We concluded that deductive impossibilities deny 100%-guarantees for security.  ...  They were first formulated in general supervised learning and optimization, which were subsequently unified through that framework.  ...  agents [42], [43] Y I No Free Lunch - supervised learning [15] Y I No Free Lunch - [16] Y I  ... 
arXiv:2109.00484v2 fatcat:466ohk3tczefdaa7xhf3i32uuu

Systems Challenges for Trustworthy Embodied Systems [article]

Harald Rueß
2022 arXiv   pre-print
A new generation of increasingly autonomous and self-learning embodied systems is about to be developed.  ...  self-integration, and continual analysis and assurance.  ...  This ability of self-learning through exploration is, of course, a far cry from supervised machine learning schemes for synthesizing, say, neural network representations for approximating functions from  ... 
arXiv:2201.03413v2 fatcat:hwprg3zjhvfuro3etecx2t4qua

Bayesian cognitive science, predictive brains, and the nativism debate

Matteo Colombo
2017 Synthese  
Unlike Connectionism, the Bayesian approach would show how unsupervised and self-supervised forms of hierarchical learning can be responsible for the quick, robust, and smooth acquisition of novel psychological  ...  These networks could successfully carry out their processing in a self-supervised fashion by learning a compact, invertible code that allowed them to reconstruct their own input on their output (i.e.,  ... 
doi:10.1007/s11229-017-1427-7 fatcat:wcjelmh47relnightn2g5n2tby

Learning Whenever Learning is Possible: Universal Learning under General Stochastic Processes [article]

Steve Hanneke
2020 arXiv   pre-print
This work initiates a general study of learning and generalization without the i.i.d. assumption, starting from first principles.  ...  We study this question in three natural learning settings: inductive, self-adaptive, and online.  ...  In the special case of an i.i.d. process, the self-adaptive setting is closely related to the semi-supervised learning setting studied in the statistical learning theory literature (Chapelle, Schölkopf  ... 
arXiv:1706.01418v2 fatcat:neq4yc5fqzdahopslcc23fgrhe

Toward an AI Physicist for Unsupervised Learning [article]

Tailin Wu
2019 arXiv   pre-print
We investigate opportunities and challenges for improving unsupervised machine learning using four common strategies with a long history in physics: divide-and-conquer, Occam's razor, unification and lifelong  ...  learning.  ...  We thank Isaac Chuang, John Peurifoy and Marin Soljačić for helpful discussions and suggestions, and the Center for Brains, Minds, and Machines (CBMM) for hospitality.  ... 
arXiv:1810.10525v3 fatcat:j75ejqkuifae5ljwvpygxy2xi4

Multi-agent Reinforcement Learning: An Overview [chapter]

Lucian Buşoniu, Robert Babuška, Bart De Schutter
2010 Studies in Computational Intelligence  
This chapter reviews a representative selection of Multi-Agent Reinforcement Learning (MARL) algorithms for fully cooperative, fully competitive, and more general (neither cooperative nor competitive)  ...  The agents must instead discover a solution on their own, using learning. A significant part of the research on multi-agent learning concerns reinforcement learning techniques.  ...  Together with the simplicity and generality of the setting, this makes RL attractive also for multi-agent learning.  ... 
doi:10.1007/978-3-642-14435-6_7 fatcat:uonoth4kijcjdohhnq73upjnzu
« Previous Showing results 1 — 15 out of 800 results