FIRAL: An Active Learning Algorithm for Multinomial Logistic Regression
This work addresses the problem of reducing labeling costs in multiclass classification for machine learning practitioners, though it is incremental as it builds on existing active learning frameworks.
The paper tackles active learning for multiclass classification using multinomial logistic regression by proving that the Fisher Information Ratio bounds excess risk and proposing an algorithm to minimize it, resulting in consistently smaller classification errors on datasets like MNIST, CIFAR-10, and 50-class ImageNet compared to five other methods.
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.