Cover Image

Sparse Autoencoders as Plug-and-Play Firewalls for Adversarial Attack Detection in VLMs

Fuente: arXiv

Saved in:

Bibliographic Details
Main Authors:	Wang, Hao, Sun, Yiqun, Wei, Pengfei, Hsieh, Lawrence B., Kawahara, Daisuke
Format:	Preprint
Published:	2026
Subjects:	Computer Vision and Pattern Recognition Artificial Intelligence Computation and Language Machine Learning
Online Access:	Acceder al recurso
Tags:	Add Tag No Tags, Be the first to tag this record!

Similar Items

Very Efficient Listwise Multimodal Reranking for Long Documents
by: Sun, Yiqun, et al.
Published: (2026)

Traveling Across Languages: Benchmarking Cross-Lingual Consistency in Multimodal LLMs
by: Wang, Hao, et al.
Published: (2025)

Return of Frustratingly Easy Unsupervised Video Domain Adaptation
by: Wei, Pengfei, et al.
Published: (2026)

Evaluating Multimodal Large Language Models on Vertically Written Japanese Text
by: Sasagawa, Keito, et al.
Published: (2025)

An Image is Worth 1/2 Tokens After Layer 2: Plug-and-Play Inference Acceleration for Large Vision-Language Models
by: Chen, Liang, et al.
Published: (2024)

XSPA: Crafting Imperceptible X-Shaped Sparse Adversarial Perturbations for Transferable Attacks on VLMs
by: Hu, Chengyin, et al.
Published: (2026)

Autoregressive Semantic Visual Reconstruction Helps VLMs Understand Better
by: Wang, Dianyi, et al.
Published: (2025)

Plug-and-Play Grounding of Reasoning in Multimodal Large Language Models
by: Chen, Jiaxing, et al.
Published: (2024)

FBHM: Functional Benchmarking and Steering of VLMs for Hateful Meme Detection
by: Bhaskar, Paramananda, et al.
Published: (2026)

Mitigating Adversarial Attacks in LLMs through Defensive Suffix Generation
by: Kim, Minkyoung, et al.
Published: (2024)

Mask the Target: A Plug-and-Play Regularizer Against LoRA Forgetting
by: Xu, Runze, et al.
Published: (2026)

Steering LVLMs via Sparse Autoencoder for Hallucination Mitigation
by: Hua, Zhenglin, et al.
Published: (2025)

Tone Matters: The Impact of Linguistic Tone on Hallucination in VLMs
by: Hong, Weihao, et al.
Published: (2026)

When VLMs Meet Image Classification: Test Sets Renovation via Missing Label Identification
by: Pang, Zirui, et al.
Published: (2025)

Have the VLMs Lost Confidence? A Study of Sycophancy in VLMs
by: Li, Shuo, et al.
Published: (2024)

Adversarial Attacks on VQA-NLE: Exposing and Alleviating Inconsistencies in Visual Question Answering Explanations
by: Yeh, Yahsin, et al.
Published: (2025)

iVISPAR -- An Interactive Visual-Spatial Reasoning Benchmark for VLMs
by: Mayer, Julius, et al.
Published: (2025)

Scaling Agentic Reinforcement Learning for Tool-Integrated Reasoning in VLMs
by: Lu, Meng, et al.
Published: (2025)

VisionFoundry: Teaching VLMs Visual Perception with Synthetic Images
by: Zhou, Guanyu, et al.
Published: (2026)

Can VLMs Recall Factual Associations From Visual References?
by: Ashok, Dhananjay, et al.
Published: (2025)

Robust Vision-Language Models via Tensor Decomposition: A Defense Against Adversarial Attacks
by: Patel, Het, et al.
Published: (2025)

X-Transfer Attacks: Towards Super Transferable Adversarial Attacks on CLIP
by: Huang, Hanxun, et al.
Published: (2025)

SPARC: Separating Perception And Reasoning Circuits for Test-time Scaling of VLMs
by: Avogaro, Niccolo, et al.
Published: (2026)

BabyVLM: Data-Efficient Pretraining of VLMs Inspired by Infant Learning
by: Wang, Shengao, et al.
Published: (2025)

GrAInS: Gradient-based Attribution for Inference-Time Steering of LLMs and VLMs
by: Nguyen, Duy, et al.
Published: (2025)

MedObvious: Exposing the Medical Moravec's Paradox in VLMs via Clinical Triage
by: Khan, Ufaq, et al.
Published: (2026)

Attention-Guided Patch-Wise Sparse Adversarial Attacks on Vision-Language-Action Models
by: Zhang, Naifu, et al.
Published: (2025)

Autoencoder-based Denoising Defense against Adversarial Attacks on Object Detection
by: Song, Min Geun, et al.
Published: (2025)

Seeing Isn't Knowing: Do VLMs Know When Not to Answer Spatial Questions (and Why)?
by: Zhang, Yue, et al.
Published: (2026)

Preserving Multi-Modal Capabilities of Pre-trained VLMs for Improving Vision-Linguistic Compositionality
by: Oh, Youngtaek, et al.
Published: (2024)

Prompt-Driven Contrastive Learning for Transferable Adversarial Attacks
by: Yang, Hunmin, et al.
Published: (2024)

Lost in Translation and Noise: A Deep Dive into the Failure Modes of VLMs on Real-World Tables
by: Singh, Anshul, et al.
Published: (2025)

WAON: A Large-Scale Japanese Image-Text Dataset for Cultural Adaptation in Contrastive Vision-Language Models
by: Sugiura, Issa, et al.
Published: (2025)

TS-P$^2$CL: Plug-and-Play Dual Contrastive Learning for Vision-Guided Medical Time Series Classification
by: Xu, Qi'ao, et al.
Published: (2025)

Quizzard@INOVA Challenge 2025 -- Track A: Plug-and-Play Technique in Interleaved Multi-Image Model
by: Cuong, Dinh Viet, et al.
Published: (2025)

Are VLMs Really Blind
by: Singh, Ayush, et al.
Published: (2024)

VLMs Can Aggregate Scattered Training Patches
by: Zhou, Zhanhui, et al.
Published: (2025)

Anatomical Region-Guided Contrastive Decoding: A Plug-and-Play Strategy for Mitigating Hallucinations in Medical VLMs
by: Liang, Xiao, et al.
Published: (2025)

PDF-WuKong: A Large Multimodal Model for Efficient Long PDF Reading with End-to-End Sparse Sampling
by: Xie, Xudong, et al.
Published: (2024)

Can World Models Benefit VLMs for World Dynamics?
by: Zhang, Kevin, et al.
Published: (2025)