AutoAL: Automated Active Learning with Differentiable Query Strategy Search

Yifeng Wang, Xueying Zhan, Siyu Huang

TL;DR: To our knowledge, we present the first automatic AL query strategy search method that can be trained in a differientiable way.

Abstract: As deep learning continues to evolve, the need for data efficiency becomes increasingly important. Considering labeling large datasets is both time-consuming and expensive, active learning (AL) provides a promising solution to this challenge by iteratively selecting the most informative subsets of examples to train deep neural networks, thereby reducing the labeling cost. However, the effectiveness of different AL algorithms can vary significantly across data scenarios, and determining which AL algorithm best fits a given task remains a challenging problem. This work presents the first differentiable AL strategy search method, named AutoAL, which is designed on top of existing AL sampling strategies. AutoAL consists of two neural nets, named SearchNet and FitNet, which are optimized concurrently under a differentiable bi-level optimization framework. For any given task, SearchNet and FitNet are iteratively co-optimized using the labeled data, learning how well a set of candidate AL algorithms perform on that task. With the optimal AL strategies identified, SearchNet selects a small subset from the unlabeled pool for querying their annotations, enabling efficient training of the task model. Experimental results demonstrate that AutoAL consistently achieves superior accuracy compared to all candidate AL algorithms and other selective AL approaches, showcasing its potential for adapting and integrating multiple existing AL methods across diverse tasks and domains.

Lay Summary:

Training deep learning models often requires labeling large amounts of data by hand, this is a both time-consuming and cost-expensive process. Active learning offers a solution by enabling the model to select the most informative examples to label next. However, the effectiveness of different selection strategies can vary across datasets and tasks, making it difficult to determine the best approach in practice.

We created AutoAL, a system that integrates multiple active learning strategies, and automatically tries multiple strategies at once and learns which combination works best for each specific dataset. Our tests on both everyday and medical images show that AutoAL consistently beats using single strategy alone. This makes it much easier and cheaper to train deep learning models, especially in areas like medical imaging where getting labeled data is particularly expensive and difficult.

Link To Code: https://github.com/haizailache999/AutoAL

Primary Area: General Machine Learning->Online Learning, Active Learning and Bandits

Keywords: Active Learning, Differentiable Bi-level Optimization

Submission Number: 11663