← Back to all news
arXiv AI By Fabian Denoodt, Sibylle Hess

When Has a Bayesian Neural Network Sampled Enough? Adaptive Inference Time with Statistical Guarantees

Read the original on arXiv AI →

The paper introduces a method that uses confidence sequences to adaptively determine the number of Monte Carlo samples needed for Bayesian neural network predictions, ensuring statistical guarantees. It applies to tasks such as class identification, full predictive distribution approximation, and probability-threshold decisions, stopping sampling once the desired confidence is achieved. Experiments demonstrate efficient computational budgeting, allocating more samples to ambiguous inputs, preserving decision reliability, and reducing overall latency compared to a fixed sample budget.

Machine-generated by The Flow from the publisher's headline and feed description — not written or checked by a human. The full article lives at arXiv AI.

Related stories

Hugging Face Trending Papers
Jul 7

Efficient Bayesian Deep Ensembles via Analytic Predictive Inference

We introduce an efficient Bayesian deep ensemble method for predictive regression designed to enhance interpretability while maintaining competitive predictive performance and computational efficiency. Our method combines the statistical rigor of Bayesian inference with the scalability of deep ensembles, providing calibrated uncertainty estimates that enable its use not only for standalone prediction but also as a component within broader learning systems.

benchmarkssafety
More like this →
arXiv Statistics ML
1d ago

Sparse Bayesian Neural Networks via Sticky Piecewise Deterministic Markov Processes

The paper introduces a method for training sparse Bayesian neural networks by integrating spike-and-slab priors into sticky piecewise deterministic Markov processes (PDMPs). It demonstrates that these modern Monte Carlo techniques can sample posterior distributions of Bayesian neural networks exactly, provided numerical rate bounds are satisfied. Experiments show that the resulting sparse models not only converge well but also exhibit greater robustness to input noise compared to their dense counterparts.

By August Arnstad, Geir Olve Storvik, Filippo Pagani, Leiv R{\o}nneberg
efficiency
More like this →