Search papers, labs, and topics across Lattice.
University of Chinese Academy of Sciences
3
0
7
Malicious instructions hidden in images can bypass existing skill scanners, exposing a critical vulnerability in LLM-based systems.
By intelligently filtering out irrelevant visual noise, TPS-Drive lets VLMs focus on what *really* matters for safe self-driving: dynamic agents, not static scenery.
State-of-the-art skeleton-based action recognition is now possible through a game-theoretic contrastive learning framework that maximizes action-relevant information while minimizing encoding redundancy.
