🏢 University of Maryland, College Park
Shadowcast: Stealthy Data Poisoning Attacks Against Vision-Language Models
·3138 words·15 mins·
loading
·
loading
Multimodal Learning
Vision-Language Models
🏢 University of Maryland, College Park
Shadowcast: A new data poisoning attack manipulates vision-language models by injecting visually similar, yet deceptively misleading, image-text pairs, causing them to generate false information.
LLM-Check: Investigating Detection of Hallucinations in Large Language Models
·2270 words·11 mins·
loading
·
loading
Natural Language Processing
Large Language Models
🏢 University of Maryland, College Park
LLM-Check efficiently detects LLM hallucinations in a single response, using internal model analysis, improving real-time applications.
Decomposing and Interpreting Image Representations via Text in ViTs Beyond CLIP
·2855 words·14 mins·
loading
·
loading
Computer Vision
Vision-Language Models
🏢 University of Maryland, College Park
This paper presents a general framework for interpreting Vision Transformer (ViT) components, mapping their contributions to CLIP space for textual interpretation, and introduces a scoring function fo…