Search papers, labs, and topics across Lattice.
Zhejiang University, Westlake University
6
0
9
Low generative perplexity in diffusion models often masks excessive repetition, with a simple fix cutting repetition to human levels while being 1.5–5x cheaper.
OPID achieves a remarkable boost in agent performance by leveraging hierarchical skills extracted from on-policy trajectories, transforming sparse rewards into dense, actionable insights.
LLMZero uncovers that adaptive training strategies can boost RL performance by up to 140% by dynamically adjusting regularization parameters in response to training dynamics.
By optimizing both tree structure and node budget, CaDDTree achieves superior token throughput without the need for offline budget searches, revolutionizing speculative decoding efficiency.
Robot path planning can be sped up by >60% while maintaining near-perfect success rates by learning to propose compact, topologically-aware search regions.
The LPCVC 2025 winning solutions showcase surprisingly effective strategies for balancing accuracy and efficiency in edge-based computer vision, pushing the boundaries of what's possible on resource-constrained devices.
