Deep Active Learning by Leveraging Training Dynamics

Abstract

Active learning theories and methods have been extensively studied in classical statistical learning settings. However, deep active learning, i.e., active learning with deep learning models, is usually based on empirical criteria without solid theoretical justification, thus suffering from heavy doubts when some of those fail to provide benefits in applications. In this paper, by exploring the connection between the generalization performance and the training dynamics, we propose a theory-driven deep active learning method (dynamicAL) which selects samples to maximize training dynamics. In particular, we prove that convergence speed of training and the generalization performance is positively correlated under the ultra-wide condition and show that maximizing the training dynamics leads to a better generalization performance. Further on, to scale up to large deep neural networks and data sets, we introduce two relaxations for the subset selection problem and reduce the time complexity from polynomial to constant. Empirical results show that dynamicAL not only outperforms the other baselines consistently but also scales well on large deep learning models. We hope our work inspires more attempts in bridging the theoretical findings of deep networks and practical impacts in deep active learning applications.

Improved Algorithms for Neural Active Learning

“…Margin algorithms are flexible and can be adapted to both streaming and pool settings. In the pool setting, a line of works utilize the neural networks in active learning to improve the empirical performance [38,44,6,18,32,47,49,56,5]. However, they do not provide performance guarantee for NN-based active learning algorithms.…”

Deep Generative Model for Multi-Class Imbalanced Learning

“…Diversity sampling algorithms label examples that are most different from each other, based on similarity metrics such as distances in penultimate layer representations (Sener & Savarese, 2017;Geifman & El-Yaniv, 2017;Citovsky et al, 2021) or discriminator networks (Gissin & Shalev-Shwartz, 2019). Lastly, gradient embeddings, which encode both softmax likelihood and penultimate layer representation, have become widely adopted in recent approaches (Ash et al, 2019;Wang et al, 2021;Elenter et al, 2022;Mohamadi et al, 2022). As an example, Ash et al ( 2019) uses a k-means++ algorithm to query a diverse set of examples in the gradient embedding space.…”