PhD Student
Department of Computing, The Hong Kong Polytechnic University
I am a PhD student at The Hong Kong Polytechnic University working on reinforcement learning with human feedback, reward modeling, reasoning in large language models, and embodied intelligence.
My broader goal is to build learning-based agents that are more reliable, adaptive, and practically useful in complex environments.
Citations
108
Since 2021 108
h-index
5
Since 2021 5
i10-index
3
Since 2021 3
Publications
11
Last synced Mar 8, 2026
Current topics: Reinforcement Learning / Reward Modeling / Large Language Model / Embodied AI. Academic profile data is synchronized from Google Scholar every three days.
Department of Computing
The Hong Kong Polytechnic University
Hong Kong
About
My research sits at the intersection of language-model alignment, reinforcement learning, and embodied intelligence. I am particularly interested in how agents can learn from feedback, reason more effectively, and behave more reliably in open-ended environments.
Across my work, I care about methods that are both rigorous and useful: clear problem formulation, reproducible experiments, and systems that can transfer beyond narrow benchmark settings.
I welcome academic collaboration, visiting opportunities, and research-oriented industry conversations related to RLHF, agent training, reasoning, and embodied AI.
Research themes
RLHF, reward modeling, reasoning in LLMs, reinforcement learning, and embodied intelligence.
Approach
Clear problem framing, reproducible experiments, and methods that transfer beyond narrow benchmarks.
Collaboration
Particularly interested in work that connects foundational research with practical agent systems and evaluation.
Updates
A compact view of recent publications and the current Google Scholar snapshot. This section is meant to give collaborators, faculty, and industry researchers a fast overview of ongoing activity.
Snapshot
Google Scholar currently lists 11 publications and 108 citations, with an h-index of 5.
Last synchronization: Mar 8, 2026.
View full Scholar profileRecent publications
2026
SonicBench: Dissecting the Physical Perception Bottleneck in Large Audio Language ModelsarXiv preprint arXiv:2601.11039, 2026
Citations
0
2025
Reasoning beyond language: A comprehensive survey on latent chain-of-thought reasoningarXiv preprint arXiv:2505.16782, 2025
Citations
39
2025
Unveiling the key factors for distilling chain-of-thought reasoningFindings of the Association for Computational Linguistics: ACL 2025, 15094-15119, 2025
Citations
37
2025
Fine-grained and multi-dimensional metrics for document-level machine translationProceedings of the 2025 Conference of the Nations of the Americas Chapter of …, 2025
Citations
7
Research
My publications span RLHF, reward modeling, reasoning in LLMs, reinforcement learning, and embodied intelligence. Citation metrics and publication metadata below are synchronized from my public Google Scholar profile.
Total citations
108
Since 2021 108
h-index
5
Since 2021 5
i10-index
3
Since 2021 3
Publications
11
Refreshed Mar 8, 2026
Selected work
arXiv 2025
X Chen, A Zhao, H Xia, X Lu, H Wang, Y Chen, W Zhang, J Wang, W Li, ...
A survey of latent chain-of-thought reasoning that maps out how reasoning can emerge beyond explicit verbalized steps, with implications for evaluation, supervision, and model design.
Citations
39
ACL Findings 2025
X Chen, Z Sun, G Wenjin, M Zhang, Y Chen, Y Sun, H Su, Y Pan, ...
An empirical study of how supervision format, reasoning granularity, and teacher quality affect the distillation of chain-of-thought reasoning into smaller language models.
Citations
37
EMNLP 2024
Y Chen, D Zhu, Y Sun, X Chen, W Zhang, X Shen
We show that more accurate reward models do not always yield better RLHF outcomes, highlighting a practical paradox between reward-model quality and downstream language-model performance.
Citations
14
NAACL 2025
Y Sun, D Zhu, Y Chen, E Xiao, X Chen, X Shen
A study of fine-grained evaluation signals for document-level machine translation, with the goal of measuring translation quality beyond coarse aggregate metrics.
Citations
7
arXiv 2024
Y Chen, X Zhang, X Wang, Z Xu, X Shen, W Zhang
We revisit Soft Actor-Critic in high-dimensional control and analyze how distribution shift in action sampling can undermine learning stability and final performance.
Citations
5
arXiv 2025
Y Chen, Y Sun, X Chen, J Wang, X Shen, W Li, W Zhang
A study of how chain-of-thought style reasoning can improve multimodal alignment in 3D vision-language settings through more structured intermediate supervision.
Citations
3
Full list
Projects
Selected repositories and reproducible research artifacts related to my publications. This section is especially useful for readers who want to evaluate implementation quality or explore follow-up work.
ACL Findings 2025
An empirical study of how supervision format, reasoning granularity, and teacher quality affect the distillation of chain-of-thought reasoning into smaller language models.
EMNLP 2024
We show that more accurate reward models do not always yield better RLHF outcomes, highlighting a practical paradox between reward-model quality and downstream language-model performance.
Teaching
Teaching is a core part of academic life. This section will be used to document course support, mentoring activities, and teaching materials as they become available.
Coming soon
Course, TA, and instructional experience will be documented here. A fuller teaching record and teaching materials page are planned.
Coming soon
Supervision, informal mentoring, reading groups, and project guidance will be added as this section grows.
Service
A standard academic homepage usually documents how the researcher contributes to the broader community, not only through papers and code, but also through service.
Coming soon
Conference reviewing, workshop organization, and community service contributions will be listed here when ready for publication.
Coming soon
Reading groups, research communities, and other forms of scholarly engagement can be added here over time.
Honors
This section will be used to record distinctions and major milestones in a format that is easy for academic and industry readers to scan.
Coming soon
Scholarships, distinctions, awards, and recognitions will be collected here as a concise record.
Coming soon
Notable research milestones, invited activities, or other professional highlights can also be summarized in this section.
CV
A dedicated CV link is planned as part of this homepage. Until a finalized public version is added, please contact me directly if you need a current CV or supporting materials.
Coming soon
A polished public CV and selected supporting materials will be linked here in a later update.
Writing
Essay
Distilling reasoning into smaller language models is not just a matter of shrinking parameters. It is a question of what kind of intermediate structure we want the student to internalize.
Essay
RLHF is often described as reward optimization, but that framing is incomplete. In practice, the real challenge is modeling the structure and limits of feedback.
Essay
Embodied AI becomes genuinely interesting when perception, action, and feedback are part of the same loop. Static world modeling alone is not enough.
Contact
I welcome academic collaboration, visiting opportunities, and research-oriented industry discussions related to RLHF, LLM reasoning, reward modeling, and embodied AI. Email is the best first contact.
Direct
yan-jun.chen@connect.polyu.hk
xzqm13143609845
Profiles