QY
Photo of Qiwei Ye

Qiwei Ye (叶启威)

Beijing Academy of Artificial Intelligence (BAAI)

Qiwei Ye (叶启威) leads the Health Computing Research Center at the Beijing Academy of Artificial Intelligence (BAAI), where he created OpenComplex. Before joining BAAI, he spent seven years at Microsoft Research Asia, where he co-created LightGBM and Suphx. His work sits at the intersection of generative AI, reinforcement learning, and AI for Science. He believes that building models capable of understanding the physical world is both the key to unlocking life science and an essential step on the road to general intelligence.

Email Google Scholar LinkedIn Twitter/X GitHub
Citations: 29,342 h-index: 17 i10-index: 23

News

  • 2026
    Introduced AREX — a family of recursively self-improving deep research agents that exploit discovery–verification asymmetry. AREX alternates between an inner research loop and an outer self-improvement loop that audits answers constraint-wise. With a 122B-A10B MoE model, it substantially outperforms comparable-scale baselines across BrowseComp, WideSearch, DeepSearchQA, and HLE. Paper Project Page HuggingFace
  • 2026
    Floyd-ARC, built on FloydNet, scores 70.5 on ARC-AGI-1 with only 154M parameters — surpassing Grok-4-thinking (1.7T) and setting a new SOTA among ARC-trained models. GitHub
  • 2026
    Introduced FloydNet — expanding Transformers to k-tensors for higher-order relational reasoning. On General TSP (N=100–200), it reaches 99.8% optimality, significantly surpassing the LKH heuristic (38.8%). Paper GitHub
  • 2025
    Launched the OpenComplex Server — OpenComplex2 (OC2) goes beyond static structure prediction, generating full conformational ensembles to capture the dynamic nature of biomolecules. Paper Project Page GitHub
  • 2025
    IneqSearch accepted at NeurIPS 2025 — solving 78.3% of 437 Olympiad-level inequalities via hybrid symbolic-neural reasoning, over 100x faster than prior SOTA. Paper
  • 2025
    UltraSelex — a single-step approach to high-affinity RNA ligand discovery, in collaboration with Heidelberg. Cover article in Nature Chemical Biology. Paper
  • 2025
    Released Sable — a pre-training paradigm that bridges sequence and structure for protein understanding. Out in Briefings in Bioinformatics. Paper GitHub
  • 2024
    "Beyond Weisfeiler-Lehman" received the ICLR 2024 Outstanding Paper Honorable Mention — a quantitative framework that rethinks how we measure GNN expressiveness. Paper GitHub

Milestones

  • 2024
    OpenComplex held the top position on CAMEO protein structure prediction for nearly 30 consecutive months.
  • 2020
    Released Suphx — the first AI to reach superhuman level in Mahjong through deep reinforcement learning.
  • 2017
    Released LightGBM at NeurIPS 2017 — now among the most widely cited machine learning papers, with 27,900+ citations.

Selected Publications

For a full list, please visit

Google Scholar
2025

Single-step Discovery of High-Affinity RNA Ligands by UltraSelex

Yaqing Zhang, Yuan Jiang, David Kuster, Qiwei Ye, Wenhao Huang, Simon Fürbacher, Jingye Zhang, Pia Doll, Wenjun Lin, Siwei Dong, Hui Wang, Zhipeng Tang, David Ibberson, Klemens Wild, Irmgard Sinning, Anthony A. Hyman, Andres Jäschke

Nature Chemical Biology 2025

New 15 citations

Research Interests

🔬

AI for Life Sciences

From protein conformational dynamics to RNA ligand discovery — using generative models and geometric learning to understand how life works at the molecular level.

🧠

Foundation Models for the Physical World

How do we build foundation models that understand 3D structure and physical interactions? Exploring this question to enable scientific discovery at the molecular and atomic scale.

⚙️

Understanding Intelligence

What does it take for machines to learn, decide, and reason? From LightGBM's efficient decision-making to Suphx's self-play under uncertainty to FloydNet's abstract reasoning — pursuing this question across different facets of intelligence.

Reinforcement LearningGraph Neural Networks3D Foundation ModelsProtein Structure PredictionMolecular Property PredictionGradient BoostingAbstract ReasoningConformational Dynamics

Experience

Tech Lead of Health Computing Research Center / Vice President

Beijing Academy of Artificial Intelligence (BAAI)

Dec 2021 – Present

Building OpenComplex for biomolecular structure prediction, and exploring the intersection of foundation models and life sciences.

Co-Director

Tsinghua AIR–BAAI Joint Research Center for Health

2021 – 2025

Senior Researcher

Microsoft Research Asia

Mar 2014 – Dec 2021

Co-created LightGBM and Suphx. Research in gradient boosting, reinforcement learning, and multi-agent systems.

M.E. in Computer Software Engineering

Peking University

2012 – 2015

Contact

chivee.ye [at] gmail.com Google Scholar LinkedIn Twitter/X GitHub (OpenComplex Lab)