Enregistré dans:
Détails bibliographiques
Auteurs principaux: Marek, Martin, Paige, Brooks, Izmailov, Pavel
Format: Preprint
Publié: 2024
Sujets:
Accès en ligne:https://arxiv.org/abs/2403.01272
Tags: Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
_version_ 1866914699922112512
author Marek, Martin
Paige, Brooks
Izmailov, Pavel
author_facet Marek, Martin
Paige, Brooks
Izmailov, Pavel
contents Benchmark datasets used for image classification tend to have very low levels of label noise. When Bayesian neural networks are trained on these datasets, they often underfit, misrepresenting the aleatoric uncertainty of the data. A common solution is to cool the posterior, which improves fit to the training data but is challenging to interpret from a Bayesian perspective. We explore whether posterior tempering can be replaced by a confidence-inducing prior distribution. First, we introduce a "DirClip" prior that is practical to sample and nearly matches the performance of a cold posterior. Second, we introduce a "confidence prior" that directly approximates a cold likelihood in the limit of decreasing temperature but cannot be easily sampled. Lastly, we provide several general insights into confidence-inducing priors, such as when they might diverge and how fine-tuning can mitigate numerical instability.
format Preprint
id arxiv_https___arxiv_org_abs_2403_01272
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Can a Confident Prior Replace a Cold Posterior?
Marek, Martin
Paige, Brooks
Izmailov, Pavel
Machine Learning
Benchmark datasets used for image classification tend to have very low levels of label noise. When Bayesian neural networks are trained on these datasets, they often underfit, misrepresenting the aleatoric uncertainty of the data. A common solution is to cool the posterior, which improves fit to the training data but is challenging to interpret from a Bayesian perspective. We explore whether posterior tempering can be replaced by a confidence-inducing prior distribution. First, we introduce a "DirClip" prior that is practical to sample and nearly matches the performance of a cold posterior. Second, we introduce a "confidence prior" that directly approximates a cold likelihood in the limit of decreasing temperature but cannot be easily sampled. Lastly, we provide several general insights into confidence-inducing priors, such as when they might diverge and how fine-tuning can mitigate numerical instability.
title Can a Confident Prior Replace a Cold Posterior?
topic Machine Learning
url https://arxiv.org/abs/2403.01272