Yizhi Li

I am currently a Computer Science PhD student funded by the University of Manchester, supervised by Prof. Chenghua Lin. I am also a co-founder of the Multimodal Art Projection (M-A-P) research community, which aims to drive open-source academia-level research to cutting-edge level as the industry. I’ve collaborated with Dr. Jie Fu and had a lot fun.
Research
My current research focus is mainly about the the post-training of LLM with reinforce learning. Specifically, as we have stepped into the new era of training LLM with one/similar recipe by setting customized goals, how to train a real generalized model under such a paradigm?
Other research questions involving post-training of LLMs and multi-modal alignment:
- How to build an effective and robust self-evolved framework for LLMs with data synthesis (maingly during post-trianing)?
- How to unifiy the understanding and generation of vision-langauge models?
- The paradgim of aligning model among the text, vision and audio modalities.
Before the LLM era, my research interests could be concluded as these topics: language model evaluation, information retrieval, fairness in NLP, music modelling, and general topics natural language modelling. More recent and detailed topics can be referred to my publication pages.
Passed Experience
- Intenrned at J.P. Morgan Artificial Intelligence Research.
- I previously worked as a research assistant at Tsinghua NLP Lab with Prof. Zhiyuan Liu.
Academic Service: reviewer at ACL, EACL, EMNLP, INLG, ISMIR, ICLR, ICASSP, NeurIPS.
news
Dec 22, 2024 | The workshop Open Science for Foundation Models (SCI-FM) is accepted at ICLR 2025, Singapore! |
---|---|
Sep 23, 2024 | Release the text, image and audio tri-modal OmniBench. |
Aug 26, 2024 | Release the comprehensive review paper Foundation Models for Music: A Survey. |
May 30, 2024 | Four papers are accepted by the ACL’24. |
May 29, 2024 | We release the fully transparent pre-trained LLM MAP-Neo and its corpus Matrix. |