Skip to content
conftrace_
2006 NeurIPS NeurIPS 2006

A PAC-Bayes Risk Bound for General Loss Functions

Abstract

We provide a PAC-Bayesian bound for the expected loss of convex combinations of classifiers under a wide class of loss functions (which includes the exponential loss and the logistic loss). Our numerical experiments with Adaboost indicate that the proposed upper bound, computed on the training set, behaves very similarly as the true loss estimated on the testing set.

🚀 Conference Pioneer - NeurIPS 2006
📈 Trend Setter - Generalization
🧭 Keyword Pioneer - logistic loss
🐣 Hot Topic Early Bird - loss function

Authors

Pascal Germain, Alexandre Lacasse, François Laviolette, Mario Marchand

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 > Bayesian & Probabilistic > Bayesian Learning Artificial Intelligence > Bayesian & Probabilistic > Bayesian Inference Machine Learning > Optimization & Theory > Generalization

Keywords

ensemble learning classification convex optimization exponential loss logistic loss pac-bayesian bound pac-bayes bound risk bound loss function convex combination classifier ensemble convex combination of classifier
Download PDF

More from NeurIPS 2006

All NeurIPS 2006 papers →
Unsupervised Regression with Applications to Nonlinear System Identification
Large Margin Hidden Markov Models for Automatic Speech Recognition
No-regret Algorithms for Online Convex Programs
Distributed Inference in Dynamical Systems
Modelling transcriptional regulation using Gaussian Processes
Sample Complexity of Policy Search with Known Dynamics