Reproduce “Adversarial Attacks in Multimodal Systems: A Practitioner's Survey” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Shashank Kapoor, Sanjay Surendranath Girija, Lakshit Arora, Dipen Pradhan, Ankit Shetgaonkar, Aman Raj
This paper presents a practitioner-focused survey of adversarial attacks targeting multimodal AI systems, which integrate various data types like text, image, video, and audio. The work categorizes attacks by type and discusses their manifestation across modalities, particularly highlighting cross-modal implications and the increasing complexity of these threats.
Key insights
- Optimization-based and backdoor attacks are the most researched attack types on multimodal systems.
- Cross-modal attacks, particularly those exploiting joint embedding spaces, are a growing concern due to their complexity and difficulty of detection.
- The threat landscape for multimodal AI is rapidly evolving, with hybrid attacks emerging, and there is a lack of integrated open-source defense tools.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.