LORE: Lagrangian-Optimized Robust Embeddings for Visual Encoders

TL;DR

LORE introduces a constrained optimization framework for adversarial fine-tuning of visual encoders, improving robustness while maintaining accuracy and enhancing interpretability in image embeddings.

Contribution

It proposes LORE, a novel unsupervised adversarial fine-tuning method using constrained optimization to balance robustness and accuracy.

Findings

Significantly improves zero-shot adversarial robustness.

Maintains high performance on clean data.

Enhances out-of-distribution generalization and interpretability.

Abstract

Visual encoders have become fundamental components in modern computer vision pipelines. However, ensuring robustness against adversarial perturbations remains a critical challenge. Recent efforts have explored both supervised and unsupervised adversarial fine-tuning strategies. We identify two key limitations in these approaches: (i) they often suffer from instability, especially during the early stages of fine-tuning, resulting in suboptimal convergence and degraded performance on clean data, and (ii) they exhibit a suboptimal trade-off between robustness and clean data accuracy, hindering the simultaneous optimization of both objectives. To overcome these challenges, we propose Lagrangian-Optimized Robust Embeddings (LORE), a novel unsupervised adversarial fine-tuning framework. LORE utilizes constrained optimization, which offers a principled approach to balancing competing goals, such as improving robustness while preserving nominal performance. By enforcing embedding-space proximity constraints, LORE effectively maintains clean data performance throughout adversarial fine-tuning. Extensive experiments show that LORE significantly improves zero-shot adversarial robustness with minimal degradation in clean data accuracy. Furthermore, we demonstrate the effectiveness of the adversarially fine-tuned CLIP image encoder in out-of-distribution generalization and enhancing the interpretability of image embeddings.