LORE: Lagrangian-Optimized Robust Embeddings for Visual Encoders

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Khodabandeh, Borna, Afzali, Amirabbas, Afsharrad, Amirhossein, Mousavi, Seyed Shahabeddin, Lall, Sanjay, Amini, Sajjad, Moosavi-Dezfooli, Seyed-Mohsen
Natura: Preprint
Pubblicazione: 2025
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866914174236360704
author Khodabandeh, Borna
Afzali, Amirabbas
Afsharrad, Amirhossein
Mousavi, Seyed Shahabeddin
Lall, Sanjay
Amini, Sajjad
Moosavi-Dezfooli, Seyed-Mohsen
author_facet Khodabandeh, Borna
Afzali, Amirabbas
Afsharrad, Amirhossein
Mousavi, Seyed Shahabeddin
Lall, Sanjay
Amini, Sajjad
Moosavi-Dezfooli, Seyed-Mohsen
contents 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.
format Preprint
id arxiv_https___arxiv_org_abs_2505_18884
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle LORE: Lagrangian-Optimized Robust Embeddings for Visual Encoders
Khodabandeh, Borna
Afzali, Amirabbas
Afsharrad, Amirhossein
Mousavi, Seyed Shahabeddin
Lall, Sanjay
Amini, Sajjad
Moosavi-Dezfooli, Seyed-Mohsen
Machine Learning
Artificial Intelligence
Computer Vision and Pattern Recognition
Optimization and Control
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.
title LORE: Lagrangian-Optimized Robust Embeddings for Visual Encoders
topic Machine Learning
Artificial Intelligence
Computer Vision and Pattern Recognition
Optimization and Control
url https://arxiv.org/abs/2505.18884