Search papers, labs, and topics across Lattice.
Affiliation:
7
0
8
0
MLLMs can leak sensitive personal information when visual evidence is lacking, but the Dynamic Relational Unlearning Framework (DRUF) significantly mitigates this risk without sacrificing performance.
Trajectories with higher generation confidence in VLAs can drive self-improvement without external rewards, leading to performance on par with oracle RL methods.
ATHENA accelerates influence computation for billion-parameter models, achieving a staggering 313.4x speedup while maintaining or improving task performance with significantly less data.
RL models trained with verifiable rewards exhibit a surprising deductive-over-abductive reasoning asymmetry, even in controlled environments, suggesting a fundamental challenge in current RLVR approaches.
Fine-grained control over reward signals unlocks significant gains in multi-trait essay scoring, outperforming standard policy optimization techniques.
Animating 4D shapes just got easier: GaussiAnimate's "Skelebones" can reanimate unseen poses with 17% better PSNR than standard methods.
By warm-starting a dynamic priority queue with seen and generated unseen visual prototypes, this CZSL method significantly mitigates distribution shift at test time, outperforming existing approaches.
