Deep Batch Active Learning by Diverse, Uncertain Gradient Lower Bounds

TL;DR

This paper introduces BADGE, a batch active learning algorithm for deep neural networks that effectively balances diversity and uncertainty in sample selection without hyperparameter tuning, outperforming other methods.

Contribution

The paper presents a novel active learning method that combines diversity and uncertainty in batch selection using gradient embeddings, without needing hyperparameter tuning.

Findings

BADGE consistently outperforms or matches existing methods across various settings.

The approach effectively balances diversity and uncertainty without hyperparameter tuning.

BADGE is versatile and applicable to different architectures and batch sizes.

Abstract

We design a new algorithm for batch active learning with deep neural network models. Our algorithm, Batch Active learning by Diverse Gradient Embeddings (BADGE), samples groups of points that are disparate and high-magnitude when represented in a hallucinated gradient space, a strategy designed to incorporate both predictive uncertainty and sample diversity into every selected batch. Crucially, BADGE trades off between diversity and uncertainty without requiring any hand-tuned hyperparameters. We show that while other approaches sometimes succeed for particular batch sizes or architectures, BADGE consistently performs as well or better, making it a versatile option for practical active learning problems.