On the convergence of loss and uncertainty-based active learning algorithms

Haimovich, Daniel, Karamshuk, Dima, Linder, Fridolin, Tax, Niek and Vojnovic, Milan ORCID: 0000-0003-1382-022X (2025) On the convergence of loss and uncertainty-based active learning algorithms. In: Globerson, A., Mackey, L., Belgrave, D., Fan, A., Paquet, U., Tomczak, J. and Zhang, C., (eds.) Advances in Neural Information Processing Systems. UNSPECIFIED. ISBN 9798331314385

Full text not available from this repository.

Abstract

Weinvestigate the convergence rates and data sample sizes required for training a machine learning model using a stochastic gradient descent (SGD) algorithm, where data points are sampled based on either their loss value or uncertainty value. These training methods are particularly relevant for active learning and data subset selection problems. For SGD with a constant step size update, we present convergence results for linear classifiers and linearly separable datasets using squared hinge loss and similar training loss functions. Additionally, we extend our analysis to more general classifiers and datasets, considering a wide range of loss-based sampling strategies and smooth convex training loss functions. Wepropose a novel algorithm called Adaptive-Weight Sampling (AWS) that utilizes SGD with an adaptive step size that achieves stochastic Polyak’s step size in expectation. We establish convergence rate results for AWS for smooth convex training loss functions. Our numerical experiments demonstrate the efficiency of AWSonvarious datasets by using either exact or estimated loss values.

Item Type:	Book Section
Official URL:	https://papers.nips.cc/paper_files/paper/2024/hash...
Additional Information:	© 2025 The Author(s)
Divisions:	Statistics
Subjects:	Q Science > QA Mathematics > QA75 Electronic computers. Computer science
Date Deposited:	13 Jun 2025 15:00
Last Modified:	13 Jun 2025 15:03
URI:	http://eprints.lse.ac.uk/id/eprint/128407

Actions (login required)

View Item