conftrace_

2006 NeurIPS NeurIPS 2006

Tighter PAC-Bayes Bounds

Abstract

This paper proposes a PAC-Bayes bound to measure the performance of Support Vector Machine (SVM) classiﬁers. The bound is based on learning a prior over the distribution of classiﬁers with a part of the training samples. Experimental work shows that this bound is tighter than the original PAC-Bayes, resulting in an enhancement of the predictive capabilities of the PAC-Bayes bound. In addition, it is shown that the use of this bound as a means to estimate the hyperparameters of the classiﬁer compares favourably with cross validation in terms of accuracy of the model, while saving a lot of computational burden.

🚀 Conference Pioneer - NeurIPS 2006

🧭 Keyword Pioneer - pac-bayes bound

🐣 Hot Topic Early Bird - generalization bound

Authors

Amiran Ambroladze, Emilio Parrado-hernández, John S. Shawe-taylor

Topics

Machine Learning > Core Methods > Classification Machine Learning > Optimization & Theory > Bayesian Inference Machine Learning > Optimization & Theory > Learning Theory Machine Learning > Optimization & Theory > Statistical Learning Machine Learning > Optimization & Theory > Theory Machine Learning > Bayesian & Probabilistic > Bayesian Learning Machine Learning > Core Methods > Support Vector Machine

Keywords

learning theory statistical learning theory hyperparameter estimation cross validation generalization error prior distribution support vector machine generalization bound pac-bayes bound hyperparameter selection machine learning theory

More from NeurIPS 2006

All NeurIPS 2006 papers →

Non-rigid point set registration: Coherent Point Drift

Learning Nonparametric Models for Probabilistic Imitation

Active learning for misspecified generalized linear models

Statistical Modeling of Images with Fields of Gaussian Scale Mixtures

Near-Uniform Sampling of Combinatorial Spaces Using XOR Constraints

Recursive Attribute Factoring