研究方向

深度学习，强化学习，科学智能，大语言模型。

学术专著

Tao Qin. Dual Learning, Springer 2020.

主要成就与荣誉

• 2017年以计算机科学家的身份荣获《北京青年》周刊 “年度匠人精神青年榜样” 奖项；

• 提出的对偶学习助力微软在2018年中英新闻翻译任务上达到了人类专家水平；

• 带领团队在WMT2019机器翻译大赛中获得8个项目的冠军；

• 2019年设计了当时最高效的语音合成模型FastSpeech，实现了百倍的加速，并成为微软云Azure服务上支持100多种语言和200多种语音的基础模型组件；

• 2019年开发了有史以来最强大的麻将AI Suphx，成为“天凤”平台上首个升至十段的AI，其稳定段位显著优于人类顶尖选手；

• 2020年在国际知名的学术出版集团施普林格·自然（Springer Nature）出版了学术专著《对偶学习》；

• 2022年发布了BioGPT模型，在生命科学领域大幅超越了其他大型语言模型，并在PubMed问答任务上首次达到了人类专家的水平；

• 荣获ICDM 2022最佳学生论文亚军。

代表性学术论文

[1]    NatureLM: Deciphering the Language of Nature for Scientific Discovery. arXiv 2025.

[2]    TamGen: drug design with target-aware molecule generation through a chemical language model. Nature Communications 2024.

[3]    HybriDNA: A Hybrid Transformer-Mamba2 Long-Range DNA Language Model. arXiv 2025.

[4]    E2Former: A Linear-time Efficient and Equivariant Transformer for Scalable Molecular Modeling. arXiv 2025.

[5]    Accelerating protein engineering with fitnesslandscape modeling and reinforcement learning. bioRxiv 2023.

[6]    BioGPT: generative pre-trained transformer for biomedical text generation and mining. Briefings in Bioinformatics 2022.

[7]    The Impact of Large Language Models on Scientific Discovery: a Preliminary Study using GPT-4. arXiv 2023.

[8]    FABind: Fast and Accurate Protein-Ligand Binding. NeurIPS 2023.

[9]    SMT-DTA: Improving Drug-Target Affinity Prediction with Semi-supervised Multi-task Training. Briefings in Bioinformatics 2023.

[10]    Pre-training Antibody Language Models for Antigen-Specific Computational Antibody Design. KDD 2023.

[11]    Dual-view Molecular Pre-training. KDD 2023.

[12]    Retrosynthetic Planning with Dual Value Networks. ICML 2023.

[13]    De Novo Molecular Generation via Connection-aware Motif Mining. ICLR 2023.

[14]    O-GNN: incorporating ring priors into molecular modeling. ICLR 2023.

[15]    R2-DDI: Relation-aware Feature Refinement for Drug-Drug Interaction Prediction. Briefings in Bioinformatics 2022.

[16]    Direct Molecular Conformation Generation. TMLR 2022.

[17]    Naturalspeech 3: Zero-shot speech synthesis with factorized codec and diffusion models. arXiv preprint 2023.

[18]    Learning to rank: from pairwise approach to listwise approach. International Conference on Machine Learning (ICML) 2007.

[19]    Fastspeech 2: Fast and high-quality end-to-end text to speech. International Conference on Learning Representations (ICLR) 2021.

[20]    MPnet: Masked and permuted pre-training for language understanding. NeurIPS 2020.

[21]    Fastspeech: Fast. robust and controllable text to speech. NeurIPS 2019.

[22]    Generalizing to unseen domains: A survey on domain generalization. IEEE Transactions on Knowledge and Data Engineering (TKDE) 2022.

[23]    Mass: Masked sequence to sequence pre-training for language generation. International Conference on Machine Learning (ICML) 2019.

[24]    Dual learning for machine translation. NeurIPS 2016.

[25]    Neural architecture optimization. NeurIPS 2018.

[26]    Achieving human parity on automatic Chinese to English news translation. arXiv preprint 2018.

[27]    LETOR: A benchmark collection for research on learning to rank for information retrieval. Information Retrieval Journal 2010.

[28]    R-drop: Regularized dropout for neural networks. NeurIPS 2021.

[29]    Incorporating BERT into neural machine translation. ICLR 2020.

[30]    A survey on neural speech synthesis. arXiv preprint 2021.

[31]    Introducing LETOR 4.0 datasets. arXiv preprint 2013.

[32]    Can generalist foundation models outcompete special-purpose tuning? Case study in medicine. arXiv preprint 2023.

[33]    An empirical study on learning to rank of tweets. ACM SIGIR 2008.

[34]    Image-to-image translation: Methods and applications. IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI) 2020.

[35]    Feature selection for ranking. European Conference on Machine Learning (ECML) 2003.

[36]    Representation degeneration problem in training natural language generation models. ACL 2020.

[37]    Naturalspeech 2: Latent diffusion models are natural and zero-shot speech and singing synthesizers. NeurIPS 2023.

[38]    Multilingual neural machine translation with knowledge distillation. ACL 2020.

[39]    NaturalSpeech: End-to-End Text-to-Speech Synthesis With Human-Level Quality. NeurIPS 2022.

[40]    Frank: a ranking method with fidelity loss. ACM SIGIR 2019.

[41]    Adaspeech: Adaptive text to speech for custom voice. Interspeech 2021.

[42]    Deliberation networks: Sequence generation beyond one-pass decoding. ACL 2021.

[43]    Understanding and improving transformer from a multi-particle dynamic system point of view. NeurIPS 2021.

[44]    Learning to teach. ICML 2017.

[45]    A study of reinforcement learning for neural machine translation. ACL 2016.

[46]    Supervised rank aggregation. ACM SIGKDD 2012.

[47]    Query dependent ranking using k-nearest neighbor. ACM SIGIR 2008.

[48]    Fully parameterized quantile function for distributional reinforcement learning. ICML 2020.