Contrastive ECOC: Learning Output Codes for Adversarial Defense

TL;DR

This paper proposes contrastive learning-based models to automatically learn output codes for ECOC, significantly improving adversarial robustness in multiclass classification tasks.

Contribution

It introduces three novel models for adaptive codebook learning in ECOC using contrastive learning, replacing manual or random codebook design.

Findings

Models outperform baselines in adversarial robustness

Superior performance across four datasets

Automated codebook learning enhances security

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.