Search papers, labs, and topics across Lattice.
8
11
12
6
Unlearning PII from VLMs is now feasible without sacrificing retained knowledge by directly manipulating attention-value representations.
Generative world models silently drop evidence and hallucinate physics when faced with conflicting real-world data, but orchestrating them via probe rollouts and constraint-injected auditing recovers physically faithful re-enactments with over 90% evidence coverage.
Bypassing retraining, OSR achieves competitive classification performance while significantly improving computational efficiency and privacy preservation.
HiCore's innovative use of multi-hypergraphs enables it to significantly reduce the bias towards popular items in conversational recommendations, achieving unprecedented fairness in user engagement.
The first comprehensive framework for understanding the security risks and defenses of mobile on-device AI systems reveals critical gaps that could jeopardize user privacy and data integrity.
MIXGUARD achieves robust privacy protection in split learning without sacrificing model utility, outperforming existing defenses against advanced data reconstruction attacks.
Uncovering critical security and privacy vulnerabilities in foundation-model-powered robots could redefine how we approach their deployment in real-world applications.
LLM agents are shockingly vulnerable to stealthy economic DoS attacks that inflate costs by up to 658x, even while returning correct answers.
