Episode-Based Active Learning with Bayesian Neural Networks

TL;DR

This paper explores active learning strategies for Bayesian neural networks in episodic data acquisition scenarios, emphasizing incremental updates and final training on accumulated data to optimize performance and reduce labeling effort.

Contribution

It introduces an effective approach combining incremental network updates with final training on accumulated data for Bayesian neural networks in episodic active learning.

Findings

Incremental updates improve model performance.

Final training on accumulated data is essential.

Reduces human labeling effort.

Abstract

We investigate different strategies for active learning with Bayesian deep neural networks. We focus our analysis on scenarios where new, unlabeled data is obtained episodically, such as commonly encountered in mobile robotics applications. An evaluation of different strategies for acquisition, updating, and final training on the CIFAR-10 dataset shows that incremental network updates with final training on the accumulated acquisition set are essential for best performance, while limiting the amount of required human labeling labor.