Yu Peng

HMPO introduces a single-stage, cost-effective reinforcement learning framework that achieves significant token compression of Chain-of-Thought reasoning with minimal loss of accuracy, applicable across various large language model architectures.

Proposed EnerInfer framework manages energy efficiency, throughput, and thermal comfort for on-device LLM inference, improving energy efficiency up to 65% without QoE violation.

This paper introduces PS-Track, a hierarchical pipeline for point-supervised Multi-Object Tracking (PS-MOT), which addresses spatial ambiguity and identity drift through Temporal-Feedback Prompting (TFP), Point-Excited Wavelet Attention (PEWA), and Uncertainty-Guided Gaussian Learning (UGL).

This paper proposes HGeo-TopoMap, a method for topological mapping in autonomous driving using an explicit prior map and implicit spatial relations, achieving substantial improvements in accuracy and robustness.