Search papers, labs, and topics across Lattice.
Huawei Foundation Model Department
3
0
5
Forget hand-coded strategies: HiSME learns how to evolve skills on the fly, leading to better agent performance and continual learning.
LLMs can reason more effectively by directly tracking their own belief in the correct answer throughout the reasoning process, enabling more targeted policy updates.
LLMs can generate significantly better software patches by first distilling issue descriptions into structured, refined requirements.
