FIRAL: An Active Learning Algorithm for Multinomial Logistic Regression

TL;DR

This paper introduces FIRAL, an active learning algorithm for multiclass logistic regression that minimizes Fisher Information Ratio to improve classification accuracy, supported by theoretical analysis and extensive experiments.

Contribution

It presents a novel active learning algorithm based on FIR minimization, with proven excess risk bounds and superior empirical performance.

Findings

FIRAL outperforms five other methods in classification error.

Theoretical bounds relate FIR to excess risk.

Experimental results on MNIST, CIFAR-10, and ImageNet support effectiveness.

Abstract

We investigate theory and algorithms for pool-based active learning for multiclass classification using multinomial logistic regression. Using finite sample analysis, we prove that the Fisher Information Ratio (FIR) lower and upper bounds the excess risk. Based on our theoretical analysis, we propose an active learning algorithm that employs regret minimization to minimize the FIR. To verify our derived excess risk bounds, we conduct experiments on synthetic datasets. Furthermore, we compare FIRAL with five other methods and found that our scheme outperforms them: it consistently produces the smallest classification error in the multiclass logistic regression setting, as demonstrated through experiments on MNIST, CIFAR-10, and 50-class ImageNet.