Efficient Semi-Supervised Adversarial Training via Latent Clustering-Based Data Reduction

TL;DR

This paper introduces latent clustering-based data reduction techniques to enhance the efficiency of semi-supervised adversarial training, significantly decreasing data and computational requirements while maintaining robustness.

Contribution

It proposes novel latent clustering methods for selecting critical data points, reducing data needs and training time in semi-supervised adversarial training.

Findings

Achieves nearly identical robustness with 5-10x less unlabeled data.

Reduces total runtime by approximately 3-4x.

Maintains strong robustness while decreasing data and computational costs.

Abstract

Learning robust models under adversarial settings is widely recognized as requiring a considerably large number of training samples. Recent work proposes semi-supervised adversarial training (SSAT), which utilizes external unlabeled or synthetically generated data and is currently the state of the art. However, SSAT requires substantial extra data to attain high robustness, resulting in prolonged training time and increased memory usage. In this paper, we propose data reduction strategies to improve the efficiency of SSAT by optimizing the amount of additional data incorporated. Specifically, we design novel latent clustering-based techniques to select or generate a small, critical subset of data samples near the model's decision boundary. While focusing on boundary-adjacent points, our methods maintain a balanced ratio between boundary and non-boundary data points, thereby avoiding overfitting. Comprehensive experiments across image benchmarks demonstrate that our methods can effectively reduce SSAT's data requirements and computational costs while preserving its strong robustness advantages. In particular, our latent-space selection scheme based on k-means clustering and our guided diffusion-based approach with LCG-KM are the most effective, achieving nearly identical robust accuracies with 5 times to 10 times less unlabeled data. When compared to full SSAT trained to convergence, our methods reduce total runtime by approximately 3 times to 4 times due to strategic prioritization of unlabeled data.