Research History¶
English中文
2026 — Year 4 (Ph.D.)¶
Doctoral Year 4
| Month | Content |
|---|---|
| 7 | 1. Patent work: Method, Apparatus, and System for Evaluating the Logical Reasoning Capability of Large Language Models 2. Completed the rebuttal for a logical reasoning study 3. Submitted an agent memory and learning study to ARR 4. Completed a draft survey on self-improving intelligence |
| 6 | 1. The Party branch was recognized as a Tsinghua University Outstanding Student Primary-Level Party Organization and Grade-A Youth League Branch; personally recognized as a Tsinghua University Outstanding Student Communist Party Member and Three-Good Student 2. Completed the first paper draft for an agent memory and learning study 3. Explored automated research and writing workflows for a survey on self-improving intelligence 4. Reviewed for NeurIPS and ARR, and participated in industry exchanges and innovation programs |
| 5 | 1. The graduate Party branch ranked 9th among 667 branches in the university-wide 2025 evaluation 2. Resubmitted a logical reasoning study 3. Completed and submitted the initial draft of a reasoning self-evolution study 4. Closed the engineering loop and added visualization capabilities to the related research system 5. Participated in exchanges and academic talks with companies, universities, and research groups |
| 4 | 1. On April 28, passed the first doctoral proposal review at the Institute of Software for Key Technologies for the Self-Evolution of Reasoning Capabilities in Foundation Models (《面向基础模型推理能力自进化的关键技术研究》) 2. Revised patent application materials and completed the rebuttal for an ICML submission 3. Advanced a reasoning self-evolution study with data synthesis, automated research, multi-machine support, experience extraction, planning, and evaluation modules 4. Studied agent harness research and practiced Vibe Coding |
| 3 | 1. Completed the first draft of a patent application 2. Completed the rebuttal for a machine-learning conference submission 3. Prepared doctoral proposal materials 4. Continued engineering development for a reasoning self-evolution study 5. Advanced an interactive motion tutoring study |
| 2 | 1. GLM-5: from Vibe Coding to Agentic Engineering was released on arXiv 2. The BDMA paper Towards Artificial Intelligence for Science: A Case Study of Using ChatGPT for Disease Causality Discovery from Biomedical Literature was formally published online and indexed in IEEE Xplore 3. Strengthened the evaluation module, timestamp mechanism, and robustness safeguards for a logical reasoning study 4. Completed the initial draft of a reasoning self-evolution study and implemented deployment and evaluation modules 5. Began preparing patent materials |
| 1 | 1. Systematically strengthened a logical reasoning study by clarifying human-intervention boundaries and stopping criteria, and by improving the evidence chain for generality, complex-logic definitions, reliability, and multi-agent related work 2. Advanced a reasoning self-evolution study by completing its framework and research plan, building logging, run management, and environment-checking infrastructure, and organizing reinforcement-learning notes and a Chinese draft 3. Prepared research-proposal materials on evidence-based medicine and LLM-assisted collaborative decision-making 4. Explored LLM applications in sports, healthcare, public safety, and multilingual settings, and attended the AGI Next Summit 5. Followed up on an interactive motion tutoring study |
2025 — Year 3 (Ph.D.)¶
- Doctoral Year 3; Main research focus: LLM research surpassing top human expert levels, focusing on large model logical reasoning capability enhancement
- Core work: Logic project (140,000+ data construction, 47+ SFT/CPT models trained), benchmark development, meta-evaluation (poetry and other artistic scenarios), o1-like model generalization evaluation
- Paper submissions: Logic work ARR→ICLR→ICML in submission; 1 collaborative paper, 1 in progress; disease causality discovery (SCI Q2) finalized; AI4Sports EDMIT paper; survey writing in progress
- Honors: Tsinghua University Bodybuilding Competition 5th place, Doctoral Academic Forum 1st place (oral + poster)
- Other: Tencent Qingyun Scholarship application, thesis proposal preparation, Wild Goose Migration Plan "Large Model-Driven Digital China Construction" project implementation (11,000-word report)
| Month | Content |
|---|---|
| 12 | 1. LRM literature review 2. ICML submission preparation 3. Bodybuilding competition 5th place |
| 11 | 1. Logic revisions; 2. Survey; 3. Algorithm research; 4. Thesis proposal preparation 1. Tencent Qingyun Scholarship application 2. Logic training 3. ICLR decision received; Logic supplementary experiments |
| 10 | 1. ICLR 2026 submission + ICLR reviewing 2. Open-source SFT+RL method reproduction, evaluation, data synthesis experiment design, training results 3. Poster and oral preparation: Doctoral Forum 1st place |
| 9 | Logic work submitted to ICLR: supplementary experiments / paper refinement |
| 8 | 1. Logic: paper revisions 2. Survey: outline; paper collection 3. Other papers: AI4Sports article EDMIT: An End-to-End Agentic Framework for Enhanced Decision-Making in Interactive Motion Tutoring Brainstorming: dLLM diffusion language model, Universal Model general large model Paper finalized: Towards Artificial Intelligence for Science: A Case Study of Using ChatGPT for Disease Causality Discovery from Biomedical Literature (SCI Q2) 4. Other: Party-building paper; Wild Goose Migration Plan: "Digital-Real Integration New Engine · Intelligent Creation · Industrial Future" — Large Model-Driven Digital China Construction project implementation, 11,000-word report + external publicity |
| 7 | 1. Benchmark work preliminarily completed |
| 6 | 1. Overall paper writing framework established; core sections draft largely completed 2. Code development core features preliminarily completed 3. 10 metadata items collected |
| 5 | 1. [Research] Benchmark coding, paper draft 2. [Paper] Towards Artificial Intelligence for Science: A Case Study of Using ChatGPT for Disease Causality Discovery — review response consideration |
| 4 | 1. Logical reasoning training: 28 SFT models, 2 CPT models 2. Revisions to 1 paper 3. 140,000 data construction 4. 2 article frameworks preliminarily constructed |
| 3 | 1. Base logic reasoning capability enhancement: 47 models trained |
| 2 | 1. Base logic reasoning capability enhancement project launched 2. Survey 3. Basic data construction 4. Basic training attempts |
| 1 | 1. o1-like model generalization evaluation and research — first month 2. LLM research surpassing top human expert levels — research direction determined 3. Meta-evaluation — poetry scenario algorithm implementation |
2024 — Year 2 (Ph.D.)¶
- Doctoral Year 2; Advisor: Jie Tang; Intern at Zhipu AI; Teaching assistant for AML & ML course; KEG Large Model Bootcamp instructor (Deep Learning Fundamentals)
- Core research: ChatGLM mathematical reasoning (PRM, full RLHF pipeline), multimodal mathematical reasoning, MalayGLM internationalization, ChatGLM mixed Chinese-English response issues, meta-evaluation (poetry and other artistic text evaluation)
- Papers and outcomes: 6 submissions (IJCAI, CogSci, ICML, etc.); 2 publications (federated learning, medical knowledge base); AiMed software copyright; Served as session chair at two paper conferences
- Honors: Challenge Cup Capital University Student Entrepreneurship Competition Gold (Beijing 1st), National 3rd; Social Practice Gold Award (2nd university-wide); Outstanding Communist Youth League Member, Computer Science Department Outstanding Student Cadre; External expert at Public Security Bureau; 2 municipal government thank-you letters
- Scholarships: Social Practice Scholarship, University Huiyan Elite Scholarship (Second Class)
- Social work: Computer Science Department Party Branch Secretary, Class Assistant; responsible for university Youth League "Tongxing" platform
- Application deployment: Public security system, medical system, LLMDailyDigest website, AML course public website (aminer.cn/aml2024)
| Paper Title | Submission Venue |
|---|---|
| ChatFUV: Chat Chain for Follow-Up Visit — Developing Personalized Follow-up Plans with Chat Chain | IJCAI AI |
| AiMed: Artificial Intelligence Large Language Model for Chinese Medicine | IJCAI AI |
| NewMed: Large Language Modeling Technology Enables Full Process Digital Intelligence in Medical Care | CogSci Cognitive Science |
| MedRad: A Reliable Assisted Decision Making Framework for Medical Large Language Models | ICML Machine Learning |
| Med-Eval: Benchmarks for the Medical Large Language Model | ICML Machine Learning |
| Doctor: The Most Reliable Digital Intelligence Healthcare Large Language Model System | - |
| OpenMonet: Open Model Orchestration Network | - |
| MedLib: Research on the Construction of a Knowledge Library for Medical Large Language Modeling | - |
| Month | Summary |
|---|---|
| 12 | KEG Large Model Bootcamp instructor — Deep Learning Fundamentals Malay LLM AML course conclusion: homework baseline, panel, grading standards, final project submission, paper session application, AML book preparation |
| 11 | Course public website: https://www.aminer.cn/aml2024 AML computing platform setup Computing Platform tutorial Meta Evaluation: Use LLM to evaluate the LLM evaluator Project proposal — poetry and other artistic text evaluation platform |
| 10 | Reinforcement Learning Survey, Self-Learning: Evaluation & Data & New Scaling Law course materials Post_Training_Scaling_Laws_Survey survey revision |
| 9 | Enhancing Mathematical Reasoning in Multimodal Large Language Models |
| 8 | Social practice summary; math literature review |
| 7 | ChatGLM mathematical reasoning| Project progress month 4: math2-prm evaluation fix; Summer doctoral required practice project |
| 6 | ChatGLM mathematical reasoning| Project progress month 3: model | PPO training; RLHF model training; model validation | PRM |
| 5 | ChatGLM mathematical reasoning| Project progress month 2: model | PRM Inference; model | PRM Training; model | PRM Evaluation |
| 4 | ChatGLM mathematical reasoning| Project progress month 1: data construction | automated step-by-step annotation; human feedback algorithm | forward auto-annotation and backward scoring feedback for process reward computation |
| 3 | 1. ChatGLM internationalization 2. Mixed Chinese-English handling |
| 2 | Personal materials preparation |
| 1 | Paper submissions × 6 |
2023 — Year 1 (Ph.D.)¶
- Research focus: medical large models and knowledge engineering
- Core outcomes: AiMed 1.0 open-source release, Doctor 1.0 deployment, Med-Eval benchmark construction launched
- Paper directions: AiMed, ChatFUV, NewMed, Med-Eval, MedRad and other medical large model work in progress
- Courses: Advanced Machine Learning (RLHF, RAG assignments and projects), CSE paper reports (KrNER, PoKG), Chinese Marxism and Contemporary
- Data construction: Preprocessed 80,000 electronic medical record entries; built guideline library and medical record library; drug instructions and lab knowledge bases
- Other: Department practice review, Zhipu AI events, Journal of Software materials, federated learning paper, etc.
| # | Task | Details |
|---|---|---|
| 1 | Model selection | Separate links |
| 2 | Knowledge base external | Similar patients batch 1: preprocessed 20,000 electronic medical records; SQL export to formatted JSON |
| 3 | Knowledge base statistics | AiMed current data |
| 4 | Knowledge base external | Guidelines |
| 5 | Department practice review | Materials preparation |
| 6 | CSE paper report | KrNER: PPT |
| 7 | CSE paper report | PoKG: PPT |
| 8 | Department practice review | On-site review |
| 9 | AiMed 1.0 release: copyright | Copyright issues |
| 10 | CSE paper report | KrNER: script preparation, video recording |
| 11 | CSE paper report | PoKG: script preparation, video recording |
| 12 | AiMed 1.0 release: service | Sensitive information filtering |
| 13 | CSE paper report | On-site presentation |
| 14 | Multi-model chain | Related research survey |
| 15 | Advanced Machine Learning | HW1 — Tokenization and compression ratio comparison (5 papers / 5 experiments) |
| 16 | AiMed 1.0 release | AiMed 1.0 project open-source |
| 17 | AiMed 1.0 release | AiMed 1.0-chat parameters release |
| 18 | AiMed 1.0 release | AiMed 1.0-paperabs parameters release |
| 19 | AiMed 1.0 release | AiMed 1.0 frontend integration |
| 20 | AiMed 1.0 release | AiMed 1.0 backend integration |
| 21 | AiMed 1.0 release: parameters | AiMed-Base full-process model parameters release |
| 22 | Chinese Marxism and Contemporary | Topic selection |
| 23 | Social work | software + Zhipu AI joint event |
| 24 | AiMed 1.0 release | Institute of Medical Information integration |
| 25 | LLM survey | |
| 26 | AiMed 2.0 | Data preparation |
| 27 | LLM | Related survey |
| 28 | Advanced Machine Learning | Project proposal |
| 29 | Advanced Machine Learning | Project proposal |
| 30 | AiMed 2.0 training | AiMed 2.0-Chat dialogue model training — round 1 |
| 31 | AiMed 2.0 training | AiMed 2.0-Chat dialogue model training — round 1 testing |
| 32 | Advanced Machine Learning | Project proposal PPT |
| 33 | Chinese Marxism and Contemporary | PPT |
| 34 | Patent | Senior Qiu patent revision |
| 35 | Advanced Machine Learning | Project proposal PPT |
| 36 | Chinese Marxism and Contemporary | PPT |
| 37 | Patent | Senior Qiu patent revision |
| 38 | Advanced Machine Learning | Project proposal PPT |
| 39 | Advanced Machine Learning | Project proposal PPT |
| 40 | Advanced Machine Learning | Project proposal PPT script |
| 41 | Institute of Medical Information report PPT | |
| 42 | Advanced Machine Learning | Project proposal PPT script |
| 43 | Advanced Machine Learning | Project discussion |
| 44 | Senior Zhang Ruilin | Journal of Software materials preparation |
| 45 | Advanced Machine Learning | Group discussion preparation |
| 46 | AiMed interface optimization | Sensitive information |
| 47 | Doctor 1.0 deployment | Model deployment to Changsha server |
| 48 | LLM | Related survey |
| 49 | Senior Zhang Ruilin | Journal of Software materials preparation |
| 50 | Advanced Machine Learning | Project discussion |
| 51 | Doctor | Guideline library and medical record library development |
| 52 | AiMed 1.0 release: service | Similar patients batch 2: preprocessed 60,000 electronic medical records; SQL export to formatted JSON |
| 53 | Engineering library and retrieval | Drug instructions, lab tests, guideline library |
| 54 | Changsha document handling | Switch to Changsha permissions for access |
| 55 | Advanced Machine Learning | Assignment 2: RLHF application in multimodal domain |
| 56 | Med-eval | Overall implementation plan |
| 57 | Med-eval | Colleague task allocation |
| 58 | Advanced Machine Learning | Poster |
| 59 | Advanced Machine Learning | PPT |
| 60 | Med-eval | Overall implementation plan |
| 61 | Med-eval | Colleague task allocation |
| 62 | Med-eval | Dataset construction: 3 datasets |
| 63 | Med-eval | Point-to-point and individual task allocation |
| 64 | Med-eval | RAG-related organization |
| 65 | Advanced Machine Learning | Assignment 3: RAG |
| 66 | Blockchain | Final project |
| 67 | Advanced Machine Learning | MedRad: paper maintenance location |
| 68 | AiMed 1.0 | AiMed: Artificial Intelligence Large Language Model for Chinese Medicine |
| 69 | ChatFUV 1.0 | ChatFUV: Chat Chain for Follow-Up Visit — Developing Personalized Follow-up Plans with Chat Chain |
| 70 | NewMed 1.0 | NewMed: Large Language Modeling Technology Enables Full Process Digital Intelligence in Medical Care |
| 71 | Med-Eval 1.0 | Med-Eval: Benchmarks for the Medical Large Language Model |
| 72 | MedRad 1.0 | MedRad: paper maintenance location |
| 73 | Federated learning paper | Journal of Software materials mailing |
2022 — Year 0 (Ph.D.)¶
- Undergraduate graduation + Ph.D. admission: Completed undergraduate thesis Research on Deep Learning Models for Chinese Electronic Medical Record Named Entity Recognition (department and university defense); entered Tsinghua University Computer Science Department for Ph.D. in September
- Research papers: MoNER (medical-oriented named entity recognition) research paper submission and multiple revisions; survey completed
- Competitions and survey: CBLUE/CBLUE2 leaderboard; BioNLP benchmark survey; CCKS/EMNLP conference analysis; blockchain key R&D indicators survey
- Summer practice: Tsinghua & Central South University medical-engineering crossover summer camp; Changsha project (AI diagnostic robot, entity recognition, disease prediction, Xiangya Hospital testing)
- Ph.D. exploration: Knowledge graph survey; patient-centered knowledge graph; data and knowledge joint-driven intelligent patient management
- Other: Personal wiki setup; exchanges with Chinese Academy of Medical Sciences / Institute of Medical Information, CAS
| Month | Week | Content |
|---|---|---|
| 1 | 1 | Undergraduate thesisResearch on Deep Learning Models for Chinese Electronic Medical Record Named Entity Recognition proposal defense |
| 2 | Personal wiki setup / BERT experiments / survey revision | |
| 3 | CBLUE Chinese medical information processing challenge leaderboard / survey revision | |
| 2 | 1 | CBLUE2 Chinese medical information processing challenge leaderboard / survey revision |
| 2 | International BioNLP benchmark survey / domestic Chinese medical information processing conferences and competitions survey / medical data source survey / survey revision / experiments | |
| 3 | Survey theoretical analysis / experiments | |
| 3 | 1 | In-depth reading of 2 papers |
| 2 | In-depth reading of 16 papers | |
| 3 | Survey completed / blockchain key R&D indicators analysis survey | |
| 4 | Blockchain key R&D indicators analysis survey 2 / experiments | |
| 4 | 1 | Undergraduate thesisResearch on Deep Learning Models for Chinese Electronic Medical Record Named Entity Recognition midterm defense |
| 2 | CCKS conference analysis / EMNLP conference analysis | |
| 3 | EMNLP detailed survey / Institute of Medical Information, Chinese Academy of Medical Sciences meeting summary / survey / experiments | |
| 4 | Survey finalized / blockchain key R&D indicators supplementary survey | |
| 5 | 1 | Research paper initial draft / medical knowledge graph survey / lab homepage survey and design |
| 2 | Lab detail page survey and design / graduation design | |
| 3 | Undergraduate thesisResearch on Deep Learning Models for Chinese Electronic Medical Record Named Entity Recognition finalized | |
| 4 | Undergraduate thesisResearch on Deep Learning Models for Chinese Electronic Medical Record Named Entity Recognition PPT | |
| 6 | 1 | Undergraduate thesisResearch on Deep Learning Models for Chinese Electronic Medical Record Named Entity Recognition defense (department and university); nested NER research |
| 2 | Chronic kidney disease full-course service system BRD v1.0 technical assessment / digital therapy insight report | |
| 3 | MoNER: A Novel Remote Supervision-based Medical-oriented Named Entity Recognition Method — research paper writing | |
| 4 | Research paper submission / survey revision / NER interface encapsulation and API documentation / lexicon update / Senior Zou work handover / Central South University summer research plan | |
| 7 | 1 | Word segmentation lexicon update / research analysis platform analysis / pre-diagnosis interface update / Central South University summer camp cooperation plan / disease prediction prior research summary / Institute of Medical Information cooperation paper directions |
| 2 | Tsinghua & Central South University summer camp medical-engineering crossover project introduction | |
| 3 | Research accumulation planning / Changsha project accumulation planning / academic journal and conference survey / Changsha work refinement / entity recognition & disease prediction & diagnostic app deployment architecture / disease-specific knowledge graph research and progress / disease prediction algorithm survey | |
| 4 | Nature Medicine journal survey / disease prediction work refinement / Changsha work plan / disease-specific knowledge graph survey | |
| 5 | Changsha environment deployment / entity recognition & disease prediction & diagnostic app deployment architecture / data flow / GPU usage / dynamic lexicon update mechanism / handover clarification / patient-centered knowledge graph construction research | |
| 8 | 1 | AI diagnostic robot: requirements analysis / overall architecture / detailed design / implementation plan |
| 2 | Named entity recognition model training / disease prediction model training / dynamic lexicon update / Changsha data familiarization | |
| 3 | Xiangya Hospital No. 1 testing / Xiangya Hospital No. 2 testing / work handover | |
| 4 | Summer league school / electronic medical record data requirements analysis / research paper revision | |
| 9 | 1 | * Basic workflow established, adjustments underway * Goals still unclear — New student orientation / research paper author response |
| 2 | Classes / Open Week report / journal survey / MoNER research paper writing | |
| 3 | Classes / knowledge graph survey | |
| 4 | Classes / Institute of Medical Information, CAS exchange | |
| 10 | 1 | Knowledge graph experiments |
| 2 | Classes / MoNER research paper revision 2 | |
| 3 | Classes / MoNER research paper overall revision plan | |
| 4 | Classes / data and knowledge joint-driven intelligent patient management | |
| 11 | 1 | Classes / MoNER research paper revision 3 / patient-centered knowledge graph survey |
| 2 | Classes / elderly renal function decline clinical assessment and early warning — NLP exchange meeting / pre-trained language model as knowledge graph research trends | |
| 12 | 1 | First half: COVID; Second half: exam period — Classes / MoNER research paper revision 4 / data and knowledge joint-driven intelligent patient management 2.0 / nested entity recognition experiments |