CORTEXA
← Browse

Xun Wu

2 papers indexed

arxivcs.LGcs.CL2026-07-20

LLM-as-a-Coach: Experiential Learning for Non-Verifiable Tasks

Tianzhu Ye, Li Dong, Guanheng Chen, He Zhu, Xun Wu, Shaohan Huang, et al.

Reinforcement learning (RL) on open-ended tasks compresses an LLM's rubric-based evaluation into a scalar reward, discarding rich textual feedback and conflating responses with distinct quality profiles. We propose Experiential Learning (EL), which repurposes the feedback model f…

View free PDFSource page
arxivcs.LG2026-06-30

FedLAB: Traceable Semantic Codebooks for Federated Multimodal Graph Foundation Learning

Zekai Chen, Kairui Yang, Xuaner Chen, Xunkai Li, Xun Wu, Rong-Hua Li, et al.

Multimodal graph foundation models aim to learn reusable knowledge from graphs enriched with text, images, attributes, and relational topology, thereby supporting diverse graph-centric and modality-centric tasks. In practice, however, such multimodal graphs are often distributed…

View free PDFSource page