IMPF-AI’s research focuses on medical artificial intelligence, large language models for traditional medicine, trustworthy agents and AI for Science. We study how AI can move beyond understanding medical knowledge and supporting decisions, toward intelligent systems that retrieve evidence, reason scientifically, discover knowledge and conduct research autonomously.
Our team works at the intersection of medicine, traditional medicine and artificial intelligence, and in recent years our research has converged on the following path:
研究方向
医学基础模型
Building specialized large language models for medicine and traditional medicine, with research on knowledge enhancement, domain instruction learning, multi-turn clinical reasoning, model evaluation, and the safety and trustworthiness of medical AI.
中医人工智能
Turning traditional medical knowledge — from unstructured classical texts, clinical records and expert experience — into computable knowledge systems, exploring TCM foundation models, knowledge graphs, multimodal data and intelligent consultation systems.
科研智能体
Research agents that autonomously perform literature search, database queries, hypothesis generation, computational analysis, evidence verification and research-report writing.
可信科学智能
Trust constraints linking models, tools, data, evidence and scientific conclusions — so that AI does not merely “generate answers” but can also explain:
代表项目
A Governed Autonomous Scientist for Traditional Chinese Medicine & Biomedicine
TCMScience is an autonomous research-agent system under development at IMPF-AI. Its goal is to take AI beyond medical question answering and turn it into an AI Scientist that can carry out research workflows.
The system separates the language model’s reasoning from independent mechanisms for trusted execution and evidence governance. Predictions, experimental results, classical-text evidence, clinical studies and database records are handled as distinct types of scientific evidence rather than simply flattened into model context.
CORE PRINCIPLE
“The model may reason. It does not define the laws of the laboratory.”
RESEARCH INFRASTRUCTURE CAPABILITIES
Expert Knowledge-Guided Language Model for Traditional Chinese Medicine
A large language model for Traditional Chinese Medicine knowledge and clinical reasoning, published in Tsinghua Science and Technology. It builds multi-task knowledge instructions around the characteristics of TCM diagnosis and treatment, and introduces:
It addresses the gaps of general-purpose LLMs in TCM — insufficient domain knowledge, unstable syndrome-differentiation reasoning, and difficulty modeling complex, multi-step treatment — and has become a foundation for IMPF-AI’s subsequent TCM model research.
Open Chinese Medical Language Models
IMPF-AI continues to open-source TCM AI models, datasets and evaluation resources. The CMLM ecosystem currently spans multiple TCM language models and data resources, including:
The goal is not a single model, but an open TCM AI infrastructure built from:
Models + Data + Knowledge + Benchmarks + Agents
AI Agent for Knowledge Discovery from Ancient Medical Literature
Knowledge discovery across large collections of classical TCM literature, moving beyond keyword search toward:
KEY RESEARCH DIRECTIONS
Its Ancient–Modern Bridge Engine explores interpretable links between classical medical concepts and modern diseases, phenotypes and biomedical knowledge.
Evaluating Medical Ethics Reasoning Across Jurisdictions
A benchmark dataset for medical ethics reasoning that examines how well large language models make ethical judgments across jurisdictions — countries and regions whose legal systems and ethical norms differ. Published at the 2026 IEEE International Conference on Digital Health (ICDH), it is part of IMPF-AI’s work on the safety and trustworthiness of medical AI.
医学知识发现
Beyond large language models and agents, the institute also works on medical data science and AI-driven knowledge discovery.
Published in Journal of Investigative Dermatology, AgenTopic is an auto-tuned topic model that analyzes large-scale psoriasis literature to identify research hotspots and their long-term evolution.
RESEARCH QUESTION
Can AI help scientists discover the structure of scientific knowledge itself?
THE SYSTEM COMBINES
AI 与科学学
IMPF-AI also studies how artificial intelligence is changing scientific research and peer review.
A study in Journal of Informetrics analyzing how textual similarity and intellectual distance between reviewers’ and authors’ research relate to peer-review behavior.
A study in Scientometrics evaluating whether large language models can judge the quality of peer-review reports, and exploring the limits of AI in future research-evaluation systems.
Together, this work extends AI research to:
AI × Scientific Discovery × Peer Review × Knowledge Production
生物医学与临床研究
IMPF-AI’s AI research is never detached from medicine itself. The team also contributes to research in traditional medicine, oncology, bioinformatics and real-world medical data.
Earlier work explored angiogenesis-related molecular signatures, tumor prognosis and immunotherapy response. This medical research background underpins a core IMPF-AI principle:
“Medical AI should be built around scientific and clinical questions — not around models alone.”
研究路径
IMPF-AI’s research is extending from domain models for medicine toward autonomous scientific research systems:
LONG-TERM GOAL · AI SCIENTIST INFRASTRUCTURE
The long-term goal is an AI Scientist Infrastructure that works alongside researchers: reading literature, connecting to databases, calling scientific tools, running computational experiments, generating hypotheses and verifying evidence — while preserving complete data provenance and clear reasoning boundaries.
代表论文
2026 IEEE International Conference on Digital Health (ICDH) (IEEE Conference)
Tsinghua Science and Technology (JCR Q1 · IF 5.4)
Journal of Investigative Dermatology (JCR Q1 · IF 7.0)
Journal of Informetrics (JCR Q1 · IF 3.9)
Scientometrics (JCR Q1 · IF 3.8)
Chinese Medicine (JCR Q1 · IF 7.4)
Frontiers in Genetics (JCR Q2 · IF 3.0)
Journal of Medical Informatics 《医学信息学杂志》 (Chinese-language journal)
Note: JCR quartiles and impact factors (IF) are from the Journal Citation Reports released in June 2026 (2025 data).
IMPF-AI explores the intersection of:
Medicine × Artificial Intelligence × Philosophy of Science × Autonomous Discovery
We care about more than “making models stronger.” We ask:
OUR DIRECTION
Building AI systems that do not merely answer scientific questions — but help investigate them.