Yue Gao

Achieve robust soccer robot control by decoupling high-level tactical decisions from low-level motor control using a hierarchical RL framework that transfers well from simulation to the real world.

Forget hand-crafted LTV pipelines: AgentLTV uses LLM-driven agents to automatically search for and evolve high-performing models, adapting to diverse data patterns and improving prediction accuracy, especially for critical high-value and negative-LTV segments.

Domain-specific knowledge hypergraphs can now be extracted with significantly improved quality by dynamically learning and applying extraction skills, outperforming static few-shot learning.