EigenShield: Causal Subspace Filtering via Random Matrix Theory for Adversarially Robust Vision-Language Models

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Darabi, Nastaran, Naik, Devashri, Tayebati, Sina, Jayasuriya, Dinithi, Krishnan, Ranganath, Trivedi, Amit Ranjan
Format: Preprint
Veröffentlicht: 2025
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866910837898215424
author Darabi, Nastaran
Naik, Devashri
Tayebati, Sina
Jayasuriya, Dinithi
Krishnan, Ranganath
Trivedi, Amit Ranjan
author_facet Darabi, Nastaran
Naik, Devashri
Tayebati, Sina
Jayasuriya, Dinithi
Krishnan, Ranganath
Trivedi, Amit Ranjan
contents Vision-Language Models (VLMs) inherit adversarial vulnerabilities of Large Language Models (LLMs), which are further exacerbated by their multimodal nature. Existing defenses, including adversarial training, input transformations, and heuristic detection, are computationally expensive, architecture-dependent, and fragile against adaptive attacks. We introduce EigenShield, an inference-time defense leveraging Random Matrix Theory to quantify adversarial disruptions in high-dimensional VLM representations. Unlike prior methods that rely on empirical heuristics, EigenShield employs the spiked covariance model to detect structured spectral deviations. Using a Robustness-based Nonconformity Score (RbNS) and quantile-based thresholding, it separates causal eigenvectors, which encode semantic information, from correlational eigenvectors that are susceptible to adversarial artifacts. By projecting embeddings onto the causal subspace, EigenShield filters adversarial noise without modifying model parameters or requiring adversarial training. This architecture-independent, attack-agnostic approach significantly reduces the attack success rate, establishing spectral analysis as a principled alternative to conventional defenses. Our results demonstrate that EigenShield consistently outperforms all existing defenses, including adversarial training, UNIGUARD, and CIDER.
format Preprint
id arxiv_https___arxiv_org_abs_2502_14976
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle EigenShield: Causal Subspace Filtering via Random Matrix Theory for Adversarially Robust Vision-Language Models
Darabi, Nastaran
Naik, Devashri
Tayebati, Sina
Jayasuriya, Dinithi
Krishnan, Ranganath
Trivedi, Amit Ranjan
Machine Learning
Cryptography and Security
Computer Vision and Pattern Recognition
Vision-Language Models (VLMs) inherit adversarial vulnerabilities of Large Language Models (LLMs), which are further exacerbated by their multimodal nature. Existing defenses, including adversarial training, input transformations, and heuristic detection, are computationally expensive, architecture-dependent, and fragile against adaptive attacks. We introduce EigenShield, an inference-time defense leveraging Random Matrix Theory to quantify adversarial disruptions in high-dimensional VLM representations. Unlike prior methods that rely on empirical heuristics, EigenShield employs the spiked covariance model to detect structured spectral deviations. Using a Robustness-based Nonconformity Score (RbNS) and quantile-based thresholding, it separates causal eigenvectors, which encode semantic information, from correlational eigenvectors that are susceptible to adversarial artifacts. By projecting embeddings onto the causal subspace, EigenShield filters adversarial noise without modifying model parameters or requiring adversarial training. This architecture-independent, attack-agnostic approach significantly reduces the attack success rate, establishing spectral analysis as a principled alternative to conventional defenses. Our results demonstrate that EigenShield consistently outperforms all existing defenses, including adversarial training, UNIGUARD, and CIDER.
title EigenShield: Causal Subspace Filtering via Random Matrix Theory for Adversarially Robust Vision-Language Models
topic Machine Learning
Cryptography and Security
Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2502.14976