Maximally Separated Active Learning

TL;DR

This work proposes an active learning method that utilizes fixed equiangular hyperspherical points as class prototypes, ensuring consistent inter-class separation and robust feature representations and introduces Maximally Separated Active Learning (MSAL) for uncertainty sampling and a combined strategy (MSAL-D) for incorporating diversity.

Abstract

Active Learning aims to optimize performance while minimizing annotation costs by selecting the most informative samples from an unlabelled pool. Traditional uncertainty sampling often leads to sampling bias by choosing similar uncertain samples. We propose an active learning method that utilizes fixed equiangular hyperspherical points as class prototypes, ensuring consistent inter-class separation and robust feature representations. Our approach introduces Maximally Separated Active Learning (MSAL) for uncertainty sampling and a combined strategy (MSAL-D) for incorporating diversity. This method eliminates the need for costly clustering steps, while maintaining diversity through hyperspherical uniformity. We demonstrate strong performance over existing active learning techniques across five benchmark datasets, highlighting the method's effectiveness and integration ease. The code is available on GitHub.

Figures (5)

• Figure 1: Overall accuracy vs. budget curves on MNIST and TinyImageNet datasets.
• Figure 2: A visualization of embeddings and selected samples for labeling. We plot the t-SNE embeddings for rounds 1,5,10,20 (left $\rightarrow$ right) for entropy sampling and our methods. Grey points are the embeddings of the unlabeled set and red points are the samples selected by the active learning strategy for that round. Our methods exhibit better separation with distinct clusters throughout the active learning process.
• Figure 3: Percentage of misclassified unlabeled samples averaged across runs on MNIST dataset for entropy vs ours.
• Figure 4: Overall accuracy vs. budget curves for $\mathbf{\beta}$ on MNIST and SVHN datasets.
• Figure 5: Ablation of AUBC and Accuracy(F-acc) vs query batch size per active learning round. We choose the batchsize with highest F-acc.