洪喆沛

School of Artificial Intelligence, South China Normal University

Foshan, Guangdong, China

[short bio] [selected papers] [Google Scholar]

email: hongzhepei@gmail.com

Language: English / 中文

I am an undergraduate student at the School of Artificial Intelligence, South China Normal University. My research interests lie in LLM post-training and agentic AI systems, with a current focus on on-policy distillation, reinforcement learning, and reliable LLM-based agents.

My research mainly spans two directions. The first is LLM post-training, including on-policy distillation, reinforcement learning, and black-box model distillation. My latest work, ROPD, explores rubric-based on-policy distillation as a black-box-compatible alternative to logit-based OPD for more sample-efficient LLM alignment.

The second is agentic AI systems, including LLM agents, multi-agent collaboration, tool use, and long-horizon task solving. I am interested in building reliable and evaluable agents that can execute complex tasks over extended interaction trajectories.

Background:

B.Eng. candidate in Software Engineering at South China Normal University, 2023-2027.
Student researcher working on LLM post-training, reinforcement learning, and agentic AI systems.

selected publications

arXiv

Rubric-based On-policy Distillation

Junfeng Fang, Zhepei Hong, Mao Zheng, and 7 more authors

arXiv preprint, May 2026

Preprint, Co-first author

Abs arXiv PDF Code

On-policy distillation (OPD) is a powerful paradigm for model alignment, yet its reliance on teacher logits restricts its application to white-box scenarios. We introduce ROPD, a rubric-based OPD framework that induces prompt-specific rubrics from teacher-student contrasts and uses them to score student rollouts for on-policy optimization. Empirically, ROPD outperforms advanced logit-based OPD methods across most scenarios and achieves up to a 10x gain in sample efficiency, positioning rubric-based OPD as a flexible, black-box-compatible alternative for scalable distillation across proprietary and open-source LLMs.
BSPC

HEAT: Hierarchical Emotion Adaptation with Progressive Thresholding for EEG Emotion and Consciousness Detection

Zhepei Hong, Rongtao Chen, Liting Li, and 3 more authors

Biomedical Signal Processing and Control, Apr 2026

Accepted, First author

PDF Code
BIBM

PR-DA: Prototype Regularization Domain Adaptation for Cross-Subject EEG-Based Emotion Recognition

Rongtao Chen, Zhepei Hong, Qi You, and 3 more authors

In IEEE International Conference on Bioinformatics and Biomedicine (BIBM), Oct 2025

Accepted, Oral presentation, Co-first author

PDF Code
TAFFC

Multi-scale Dynamic Temporal Network with Graph Matching Domain Adaptation for Cross-Subject EEG Emotion Recognition

Rongtao Chen, Zhepei Hong, Liting Li, and 3 more authors

IEEE Transactions on Affective Computing, Mar 2026

Accepted, Co-first author

PDF Code