Deep Batch Active Learning by Diverse, Uncertain Gradient Lower Bounds

Jordan T. Ash, Chicheng Zhang, Akshay Krishnamurthy, John Langford, Alekh Agarwal
6/9/2019
cs.LGstat.ML

Official code: JordanAsh/badge (217 stars, last commit 2024-06-05).

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.

View on arXivView PDF

Code Implementations(1)

JordanAsh/badgeOfficial100%

An implementation of the BADGE batch active learning algorithm.

21737PythonMar 24, 20202 years ago

Similar Papers

Discriminative Active Learning

88% match1 repo2052019

Active Learning for Convolutional Neural Networks: A Core-Set Approach

84% match1 repo2822017

Bayesian Batch Active Learning as Sparse Subset Approximation

84% match1 repo402019

Class Balanced Dynamic Acquisition for Domain Adaptive Semantic Segmentation using Active Learning

84% match2023

Streaming Active Learning with Deep Neural Networks

84% match1 repo142023

DIVEBATCH: Accelerating Model Training Through Gradient-Diversity Aware Batch Size Adaptation

82% match2025

Cite this paper

@article{ash2019deep,
  title  = {Deep Batch Active Learning by Diverse, Uncertain Gradient Lower Bounds},
  author = {Jordan T. Ash and Chicheng Zhang and Akshay Krishnamurthy and John Langford and Alekh Agarwal},
  year   = {2019},
  eprint = {1906.03671},
  archivePrefix = {arXiv},
  url    = {https://arxiv.org/abs/1906.03671}
}

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/1906.03671.svg)](https://codeofpaper.com/paper/1906.03671)

Discussion