[Paper Review] A Survey of Deep Active Learning

Pengzhen Ren, Xiao Yun+6 more|arXiv (Cornell University)|Aug 30, 2020

Machine Learning and Algorithms39 citations

TL;DR

This survey provides a comprehensive classification and overview of DeepAL, detailing challenges, query strategies, architectures, applications, and future directions.

ABSTRACT

Active learning (AL) attempts to maximize the performance gain of the model by marking the fewest samples. Deep learning (DL) is greedy for data and requires a large amount of data supply to optimize massive parameters, so that the model learns how to extract high-quality features. In recent years, due to the rapid development of internet technology, we are in an era of information torrents and we have massive amounts of data. In this way, DL has aroused strong interest of researchers and has been rapidly developed. Compared with DL, researchers have relatively low interest in AL. This is mainly because before the rise of DL, traditional machine learning requires relatively few labeled samples. Therefore, early AL is difficult to reflect the value it deserves. Although DL has made breakthroughs in various fields, most of this success is due to the publicity of the large number of existing annotation datasets. However, the acquisition of a large number of high-quality annotated datasets consumes a lot of manpower, which is not allowed in some fields that require high expertise, especially in the fields of speech recognition, information extraction, medical images, etc. Therefore, AL has gradually received due attention. A natural idea is whether AL can be used to reduce the cost of sample annotations, while retaining the powerful learning capabilities of DL. Therefore, deep active learning (DAL) has emerged. Although the related research has been quite abundant, it lacks a comprehensive survey of DAL. This article is to fill this gap, we provide a formal classification method for the existing work, and a comprehensive and systematic overview. In addition, we also analyzed and summarized the development of DAL from the perspective of application. Finally, we discussed the confusion and problems in DAL, and gave some possible development directions for DAL.

Motivation & Objective

• Explain why combining deep learning (DL) and active learning (AL) is valuable for reducing labeling costs in high-dimensional data.
• Provide a formal framework and taxonomy for DeepAL methods and strategies.
• Summarize existing DeepAL applications across fields to identify common patterns and gaps.
• Highlight challenges in integrating DL with AL and propose directions for future research.

Proposed method

• Define the pool-based DeepAL framework and formalize the optimization goal for reducing labeled data while achieving target accuracy.
• Classify DeepAL query strategies into Batch Mode DeepAL (BMDAL), uncertainty-based and hybrid strategies, deep Bayesian active learning (DBAL), density-based methods, and automated DeepAL design.
• Discuss how to address DL-specific challenges such as model uncertainty, data scarcity for labeled samples, and pipeline inconsistency via Bayesian methods, data augmentation, pseudo-labels, and semi-supervised training.
• Describe batch acquisition strategies like BALD and BatchBALD, and contrast uncertainty-only vs. diversity-aware approaches (e.g., Exploration-P, DMBAL, BADGE, WAAL, TA-VAAL).
• Outline two core DeepAL components: (i) the AL query strategy on unlabeled data, and (ii) the DL model training and optimization process.

(a) Structure diagram of convolutional neural network.

Experimental results

Research questions

• RQ1 What are the key challenges in applying traditional AL strategies to DL models?
• RQ2 How can batch-based query strategies be designed to balance information content and sample diversity in DeepAL?
• RQ3 What methods (Bayesian, semi-supervised, augmentation) help DL-based AL cope with limited labeled data and high uncertainty?
• RQ4 How can DeepAL architectures and query strategies be automated or task-aware to improve performance across domains?
• RQ5 What are the main application areas and future directions for DeepAL?

Key findings

• Batch-based querying (BatchBALD) improves over one-by-one BALD by accounting for inter-sample correlations.
• Uncertainty-based strategies are simple and effective for small batches but can harm diversity and distribution coverage.
• Hybrid strategies that balance uncertainty and diversity (e.g., BADGE, WAAL, TA-VAAL) show strong practical promise without heavy hyperparameter tuning.
• Bayesian deep learning and data augmentation help mitigate overconfidence and data scarcity in DL-based AL.
• There is a need for batch-query design that scales to DL, handles high-dimensional data, and aligns with end-task performance across applications.

(b) The pool-based active learning cycle.