Contrastive ECOC: Learning Output Codes for Adversarial Defense

2025-08-14

AI Summary

Traditional ECOC methods rely on handcrafted or random codebook design, suffering from poor generalization and limited data adaptivity, thereby failing to enhance adversarial robustness. To address this, we propose CL-ECOC—the first end-to-end differentiable ECOC framework grounded in contrastive learning—that automatically learns data-adaptive, semantically discriminative, and error-correcting codebooks. Our key innovation lies in embedding inter-class contrastive constraints directly into the ECOC encoding space, enabling joint optimization of encoding representations and classifiers, which significantly improves robustness against adversarial attacks. Extensive experiments on four benchmark datasets demonstrate that CL-ECOC consistently outperforms classical ECOC variants and state-of-the-art adversarial training baselines. The source code is publicly available.

Technology Category

• Machine Learning: Adversarial Learning & Robustness
• Computer Vision: Adversarial Attacks & Robustness
• Game Theory and Economic Paradigms: Adversarial Learning

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Economics, Online Markets and Human Computation: Data quality aspects of human-annotated datasets
• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems

Abstract

Although one-hot encoding is commonly used for multiclass classification, it is not always the most effective encoding mechanism. Error Correcting Output Codes (ECOC) address multiclass classification by mapping each class to a unique codeword used as a label. Traditional ECOC methods rely on manually designed or randomly generated codebooks, which are labor-intensive and may yield suboptimal, dataset-agnostic results. This paper introduces three models for automated codebook learning based on contrastive learning, allowing codebooks to be learned directly and adaptively from data. Across four datasets, our proposed models demonstrate superior robustness to adversarial attacks compared to two baselines. The source is available at https://github.com/YuChou20/Automated-Codebook-Learning-with-Error-Correcting-Output-Code-Technique.

Problem

• Improving multiclass classification with adaptive codebook learning
• Enhancing adversarial robustness via contrastive ECOC models
• Automating codebook generation to replace manual or random designs

Innovation

• Automated codebook learning via contrastive learning
• Adaptive ECOC codewords from data
• Enhanced robustness against adversarial attacks