Walking on the Fiber: A Simple Geometric Approximation for Bayesian Neural Networks

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Reichlin, Alfredo, Vasco, Miguel, Kragic, Danica
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866908683901861888
author Reichlin, Alfredo
Vasco, Miguel
Kragic, Danica
author_facet Reichlin, Alfredo
Vasco, Miguel
Kragic, Danica
contents Bayesian Neural Networks provide a principled framework for uncertainty quantification by modeling the posterior distribution of network parameters. However, exact posterior inference is computationally intractable, and widely used approximations like the Laplace method struggle with scalability and posterior accuracy in modern deep networks. In this work, we revisit sampling techniques for posterior exploration, proposing a simple variation tailored to efficiently sample from the posterior in over-parameterized networks by leveraging the low-dimensional structure of loss minima. Building on this, we introduce a model that learns a deformation of the parameter space, enabling rapid posterior sampling without requiring iterative methods. Empirical results demonstrate that our approach achieves competitive posterior approximations with improved scalability compared to recent refinement techniques. These contributions provide a practical alternative for Bayesian inference in deep learning.
format Preprint
id arxiv_https___arxiv_org_abs_2512_01500
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Walking on the Fiber: A Simple Geometric Approximation for Bayesian Neural Networks
Reichlin, Alfredo
Vasco, Miguel
Kragic, Danica
Machine Learning
Bayesian Neural Networks provide a principled framework for uncertainty quantification by modeling the posterior distribution of network parameters. However, exact posterior inference is computationally intractable, and widely used approximations like the Laplace method struggle with scalability and posterior accuracy in modern deep networks. In this work, we revisit sampling techniques for posterior exploration, proposing a simple variation tailored to efficiently sample from the posterior in over-parameterized networks by leveraging the low-dimensional structure of loss minima. Building on this, we introduce a model that learns a deformation of the parameter space, enabling rapid posterior sampling without requiring iterative methods. Empirical results demonstrate that our approach achieves competitive posterior approximations with improved scalability compared to recent refinement techniques. These contributions provide a practical alternative for Bayesian inference in deep learning.
title Walking on the Fiber: A Simple Geometric Approximation for Bayesian Neural Networks
topic Machine Learning
url https://arxiv.org/abs/2512.01500