Search papers, labs, and topics across Lattice.
3
0
5
0
STAMP redefines credit assignment in deep search agents, leading to up to 5.5-point gains in performance by effectively linking actions to their supporting evidence.
SCOPE-RL boosts LLM accuracy by over 11% while cutting reasoning tokens significantly, reshaping how we optimize reasoning paths in AI.
DipGuava achieves photorealistic and personalized 3D head avatars from monocular video by explicitly disentangling facial appearance, outperforming prior methods in visual quality and quantitative performance.
