Prediction-Oriented Bayesian Active Learning

Freddie Bickford Smith, Andreas Kirsch, Sebastian Farquhar, Yarin Gal, Adam Foster +1 more
4/17/2023
cs.LGstat.ML

No official code found. Best implementation: fbickfordsmith/epig (41 stars, MIT, last commit 2025-09-04).

Abstract

Information-theoretic approaches to active learning have traditionally focused on maximising the information gathered about the model parameters, most commonly by optimising the BALD score. We highlight that this can be suboptimal from the perspective of predictive performance. For example, BALD lacks a notion of an input distribution and so is prone to prioritise data of limited relevance. To address this we propose the expected predictive information gain (EPIG), an acquisition function that measures information gain in the space of predictions rather than parameters. We find that using EPIG leads to stronger predictive performance compared with BALD across a range of datasets and models, and thus provides an appealing drop-in replacement.

View on arXivView PDF

Code Implementations(1)

fbickfordsmith/epigCommunity100%
MIT

Bayesian active learning with EPIG data acquisition

418Shell, PythonFeb 20, 20231 years agoMIT

Similar Papers

Making Better Use of Unlabelled Data in Bayesian Active Learning

84% match1 repo412024

Big Batch Bayesian Active Learning by Considering Predictive Probabilities

83% match2025

Myopic Bayesian Decision Theory for Batch Active Learning with Partial Batch Label Sampling

81% match1 repo52025

Active Statistical Inference

80% match1 repo192024

Understanding Uncertainty-based Active Learning Under Model Mismatch

80% match2024

Deep Bayesian Active Learning for Preference Modeling in Large Language Models

80% match1 repo72024

Cite this paper

@article{smith2023predictionoriented,
  title  = {Prediction-Oriented Bayesian Active Learning},
  author = {Freddie Bickford Smith and Andreas Kirsch and Sebastian Farquhar and Yarin Gal and Adam Foster and Tom Rainforth},
  year   = {2023},
  eprint = {2304.08151},
  archivePrefix = {arXiv},
  url    = {http://arxiv.org/abs/2304.08151v1}
}

Add a badge to your README

Shows whether official code is linked for this paper.

Code of Paper badge
[![Code of Paper](https://codeofpaper.com/badge/2304.08151.svg)](https://codeofpaper.com/paper/2304.08151)

Discussion