Filters








193 Hits in 3.6 sec

A PAC-Bayes Bound for Tailored Density Estimation [chapter]

Matthew Higgs, John Shawe-Taylor
2010 Lecture Notes in Computer Science  
In this paper we construct a general method for reporting on the accuracy of density estimation.  ...  Using variational methods from statistical learning theory we derive a PAC, algorithm-dependent bound on the distance between the data generating distribution and a learned approximation.  ...  Conclusions and Future Work In this paper we have derived a PAC-Bayes bound for density estimation with a loss function that allows us to tailor the density estimate to a function class of interest.  ... 
doi:10.1007/978-3-642-16108-7_15 fatcat:c745dpjybjdllkqhakzivtmsra

A Primer on PAC-Bayesian Learning [article]

Benjamin Guedj
2019 arXiv   pre-print
The present paper aims at providing a self-contained survey on the resulting PAC-Bayes framework and some of its main theoretical and algorithmic developments.  ...  Generalised Bayesian learning algorithms are increasingly popular in machine learning, due to their PAC generalisation properties and flexibility.  ...  The author warmly thanks Omar Rivasplata for his careful reading and suggestions.  ... 
arXiv:1901.05353v3 fatcat:vy73fwwanvfofbp3azhsrdq5v4

Risk bounds for aggregated shallow neural networks using Gaussian prior [article]

Laura Tinsi, Arnak S. Dalalyan
2022 arXiv   pre-print
The main contribution is a precise nonasymptotic assessment of the estimation error appearing in the PAC-Bayes bound.  ...  Combining bounds on estimation and approximation errors, we establish risk bounds that are sharp enough to lead to minimax rates of estimation over Sobolev smoothness classes.  ...  We say then that f n satisfies a PAC-Bayes inequality in-expectation.  ... 
arXiv:2112.11086v2 fatcat:4vujmvoiffdivnuxbyn3sfa4u4

Bayesian fractional posteriors [article]

Anirban Bhattacharya, Debdeep Pati, Yun Yang
2016 arXiv   pre-print
Second, we derive a novel Bayesian oracle inequality based on a PAC-Bayes inequality in misspecified models.  ...  We also illustrate the theory in Gaussian process regression and density estimation problems.  ...  Many previous results on PAC-Bayes type inequalities are specifically tailored to classification (bounded loss, [11, 12, 51] ) or regression (squared loss, [16, 27, 39, 51] ) problems.  ... 
arXiv:1611.01125v2 fatcat:ucscj4we7ja55hvyrbkdy5efgq

Meta-Learning by Adjusting Priors Based on Extended PAC-Bayes Theory [article]

Ron Amit, Ron Meir
2019 arXiv   pre-print
We present a framework for meta-learning that is based on generalization error bounds, allowing us to extend various PAC-Bayes bounds to meta-learning.  ...  Learning takes place through the construction of a distribution over hypotheses based on the observed tasks, and its utilization for learning a new task.  ...  ACKNOWLEDGMENTS We thank Asaf Cassel, Guy Tennenholtz, Baruch Epstein, Daniel Soudry, Elad Hoffer and Tom Zahavy for helpful discussions of this work, and the anonymous reviewers for their helpful comment  ... 
arXiv:1711.01244v8 fatcat:uvbegjw6erezjebbinjrrejzpe

PAC-Bayesian Generalisation Error Bounds for Gaussian Process Classification

Matthias Seeger, Peter Bartlett
2003 Journal of machine learning research  
In this paper, by applying the PAC-Bayesian theorem of McAllester (1999a), we prove distributionfree generalisation error bounds for a wide range of approximate Bayesian GP classification techniques.  ...  As is shown in experiments on a real-world task, the bounds can be very tight for moderate training sample sizes.  ...  PAC-Bayesian bound for Bayes classifiers Very recently, Meir and Zhang (2003) obtained a strong PAC-Bayesian result for Bayes classifiers which can be written as expectations over a uniformly bounded  ... 
doi:10.1162/153244303765208386 fatcat:apqdzkzodjfwphsfjbm7mcw3ge

Editors' Introduction [chapter]

Sanjay Jain, Rémi Munos, Frank Stephan, Thomas Zeugmann
2013 Lecture Notes in Computer Science  
He formulates a general framework for a large class of learning algorithms for such languages and, using this framework, he reviews Angluin's classical LSTAR algorithm and compares it with various contemporary  ...  He also studied the learnability of regular languages and context-free languages; a sample result, obtained in collaboration with Franck Thollard, is that the class of regular languages can be PAC-learned  ...  In A PAC-Bayes Bound for Tailored Density Estimation, Matthew Higgs and John Shawe-Taylor consider the problem of density estimation with an unusual twist: they want their solution to be tailored to the  ... 
doi:10.1007/978-3-642-40935-6_1 fatcat:pchrsvhjezfbvh6dfplqhxhgcy

On some recent advances on high dimensional Bayesian statistics

Nicolas Chopin, Sébastien Gadat, Benjamin Guedj, Arnaud Guyader, Elodie Vernet, Aurélien Garivier
2015 ESAIM Proceedings and Surveys  
On the theoretical side, we describe some recent advances in Bayesian consistency for a nonparametric hidden Markov model as well as new pac-Bayesian results for different models of high dimensional regression  ...  After giving some brief motivations in a short introduction, we describe new advances in the understanding of Bayes posterior computation as well as theoretical contributions in non parametric and high  ...  The pac-Bayesian paradigm The pac theory consists in deriving risk bound on randomized estimators (see for example [Val84] ).  ... 
doi:10.1051/proc/201551016 fatcat:ydqnd43mlrgk5je4hh7ywvfwd4

PAC-Bayes and Domain Adaptation [article]

Pascal Germain, François Laviolette, Emilie Morvant
2018 arXiv   pre-print
a new tighter domain adaptation bound for the target risk.  ...  We provide two main contributions in PAC-Bayesian theory for domain adaptation where the objective is to learn, from a source distribution, a well-performing majority vote on a different, but related,  ...  In this scenario, one may estimate the values of β q (T X S X ), and even η T \S , by using unsupervised density estimation methods.  ... 
arXiv:1707.05712v2 fatcat:sgvyai2sczavhb2kxda2qsoosi

Information-Theoretic Local Minima Characterization and Regularization [article]

Zhiwei Jia, Hao Su
2020 arXiv   pre-print
We provide theoretical analysis including a generalization bound and empirically demonstrate the success of our approach in both capturing and improving the generalizability of DNNs.  ...  Experiments are performed on CIFAR-10, CIFAR-100 and ImageNet for various network architectures.  ...  We provide its theoretical analysis, primarily a generalization bound based on PAC-Bayes (McAllester, 1999b; a) .  ... 
arXiv:1911.08192v2 fatcat:7ufib6jxrjejpjpr5tas3wadoi

10.1162/153244301753683717

2000 Applied Physics Letters  
We suggest the Bayes point machine as a well-founded improvement which approximates the Bayes-optimal decision by the centre of mass of version space.  ...  Kernel-classifiers comprise a powerful class of non-linear decision functions for binary classification.  ...  Special thanks go to Patrick Haffner for pointing out the speed improvement by exploiting sparsity of the MNIST and USPS images and to Jun Liao for pointing out a mistake in Algorithm 2.  ... 
doi:10.1162/153244301753683717 fatcat:v6rvmmko2ffm5m4cu3wnkw3dyy

Pessimistic Model-based Offline Reinforcement Learning under Partial Coverage [article]

Masatoshi Uehara, Wen Sun
2021 arXiv   pre-print
., realizability in the function class), CPPO has a PAC guarantee with offline data only providing partial coverage, i.e., it can learn a policy that competes against any policy that is covered by the  ...  ; (2) factored MDP where the partial coverage condition is defined using density ratio based concentrability coefficients associated with individual factors.  ...  Acknowledgement The authors would like to thank Nan Jiang, Tengyang Xie for valuable feedback.  ... 
arXiv:2107.06226v2 fatcat:samqjfn7crgmxaeqhf3yx3snbi

PAC-Bayes and Domain Adaptation

Pascal Germain, Amaury Habrard, Francois Laviolette, Emilie Morvant
2019 Neurocomputing  
domain adaptation bound for the target risk.  ...  We provide two main contributions in PAC-Bayesian theory for domain adaptation where the objective is to learn, from a source distribution, a well-performing majority vote on a different, but related,  ...  In this scenario, one may estimate the values of β q (T X S X ), and even η T \S , by using unsupervised density estimation methods.  ... 
doi:10.1016/j.neucom.2019.10.105 fatcat:kxc2yfvawvhnnnef5bdwxxlif4

Rademacher Complexity Bounds for a Penalized Multi-class Semi-supervised Algorithm (Extended Abstract)

Yury Maximov, Massih-Reza Amini, Zaid Harchaoui
2018 Proceedings of the Twenty-Seventh International Joint Conference on Artificial Intelligence  
We propose Rademacher complexity bounds for multi-class classifiers trained with a two-step semi-supervised model.  ...  fixed threshold stands for clustering consistency.  ...  On another level and under the PAC-Bayes setting, [Kääriäinen, 2005] showed that in the realizable case where the hypothesis set contains the Bayes classifier, the obtained excess risk bound takes the  ... 
doi:10.24963/ijcai.2018/800 dblp:conf/ijcai/MaximovAH18 fatcat:ey5qlhwnsfafrfyxqy447wu4o4

On Generalization Error Bounds of Noisy Gradient Methods for Non-Convex Learning [article]

Jian Li, Xuanyuan Luo, Mingda Qiao
2020 arXiv   pre-print
We develop a new framework, termed Bayes-Stability, for proving algorithm-dependent generalization error bounds.  ...  We obtain new generalization bounds for the continuous Langevin dynamic in this setting by developing a new Log-Sobolev inequality for the parameter distribution at any time.  ...  We develop a new method for proving generalization bounds, termed as Bayes-Stability, by incorporating ideas from the PAC-Bayesian theory into the stability framework.  ... 
arXiv:1902.00621v4 fatcat:4jqqzwnq6ral3lci4zmmjbtutm
« Previous Showing results 1 — 15 out of 193 results