Professor Zhou’s research focuses on fundamental questions in artificial intelligence. He has long been engaged in foundational AI theory and core frontier technologies, with a focus on on natural language processing, deep learning, and artificial general intelligence. He was among the researchers who developed a structured self-attentive method for sentence representation, as well as widely cited neural models for 

abstractive and extractive text summarization. His work has contributed to the advancement of natural language representation, understanding, and generation.

His research spans artificial general intelligence (AGI), large language models, deep learning, natural language processing, AI safety, AI alignment, and AI for Science.

Across these areas, he has conducted systematic investigations ，builting a substantial body of research with both academic and real-world relevance.

Selected Publications & Open-Source Models：

On the Nature of Intelligence

[1]  “A Structured Self-Attentive Sentence  Embedding.” ICLR, 2017.

Zhouhan Lin, Minwei Feng, Cicero Nogueira dos Santos, Mo Yu, Bing Xiang, Bowen Zhou†, Yoshua Bengio^†.

• Pioneering First task-agnostic self-attention and multi-head mechanisms — adopted by the Transformer,  the foundation of GPT, BERT, and all modern LLMs.

[2] “Abstractive Text Summarization Using Sequence-to-Sequence RNNs and Beyond.” CoNLL, 2016.

Ramesh Nallapati, Bowen Zhou^†, Cicero Nogueira dos Santos, Caglar Gulcehre, Bing Xiang.

• Seminal First attention-based seq2seq for abstractive summarization; established the paradigm for modern AIGC text generation.

[3] “SummaRuNNer: A Recurrent Neural Network Based Sequence Model for Extractive Summarization of Documents.” AAAI, 2017.

Ramesh Nallapati, Feifei Zhai, Bowen Zhou^†.

• Foundational Introduced a neural sequence model for extractive summarization; foundational for neural document understanding.

[4]  “Process Reinforcement Through Implicit Rewards.” Transactions on Machine Learning Research (TMLR), 2026.

Ganqu Cui, Lifan Yuan, Zefan Wang, Bowen Zhou^†, Ning Ding, et al.

• RL Theory Process-level reinforcement learning with implicit reward shaping for reasoning.

[5] “The Entropy Mechanism of Reinforcement Learning for Reasoning Language Models.” ICLR, 2026.

Ganqu Cui, Yuchen Zhang, Jiacheng Chen, Bowen Zhou^†, Ning Ding, et al.

• RL Theory Theoretical analysis of entropy’s role in RL training dynamics for reasoning models.

AI in Practice

[6] “Integrating Speech Biomarkers and Large Language Models for Adolescent Suicide Risk Detection with Mobile Application for Real-World Evaluation.” Cell Reports Medicine, 2026.

Chang Lei, Ziyun Cui, Yinan Duan, Zhijun Wu, Diyang Qu, Wen Wu, et al.; Bowen Zhou^†.

• AI for Health Multimodal AI for mental health screening via speech analysis and LLMs.

[7]  “TTRL: Test-Time Reinforcement Learning.” NeurIPS, 2025.

Yuxin Zuo, Kaiyan Zhang, Shang Qu, Bowen Zhou^†, et al.

• RL Reasoning Enabling LLMs to learn and improve reasoning strategies at test time without additional training data.

[8] “Intern-VL Series: Open-Source Multimodal Foundation Models.” Open-source models, 2022–2025.

Shanghai AI Lab Intern-VL Team (led by Bowen Zhou^† ).

• Multimodal foundation models A family of open-source multimodal models from Intern-VL to InternVL3.5, ranking in the first echelon of open-source models for cross-modal understanding and reasoning.

Safety & Trustworthiness

[9] “Frontier AI Risk Management Framework in Practice: A Risk Analysis Technical Report.” arXiv:2507.16534, 2025.

Shanghai AI Lab Center for Safe & Trustworthy AI（Xiaoyang Chen, Chaochao Lu, Jing Shao, Lewen Yan, et al.; incl. Bowen Zhou）。

• Frontier risk framework A comprehensive assessment of frontier AI risks across seven domains, guided by the “AI-45ř Law” with red/yellow/green risk zone classification. Adopted by international AI safety institutions and cited in follow-on research on AI risk evaluation and governance.

[10]  “Reducing Human Priors in Scalable Formal Software Verification with RL in LLMs — A Preliminary Study on Dafny.” ICLR, 2026.

Chuanhao Yan, Fengdi Che, Xuhan Huang, et al., Bowen Zhou^†, Jie Fu^†.

• Formal Verification RL-guided LLMs for automated formal verification of software correctness.

[11]  “Trustworthy AI: From Principles to Practices.” ACM Computing Surveys, 2023.

Bo Li, Peng Qi, Bo Liu, Shuai Di, Jingen Liu, Jiquan Pei, Jinfeng Yi, Bowen Zhou^† .

• Comprehensive survey Covering fairness, robustness, explainability, and privacy in AI systems.

[12]  “Towards AI-45° Law: A Roadmap to Trustworthy AGI.” 2024.

Chao Yang, Chaochao Lu, Yingchun Wang, Bowen Zhou^† .

• AGI safety Proposed the “AI-45-Degree Law” for balancing capability with safety.

AGI for Science and Beyond

[13] “Building AGI through Specialized Generalist AI: pathways and key issues.” Communications of CCF, 2024.

Bowen Zhou^† .

• AGI architecture Introduced SGAI framework and SAGE (Synergistic AGI Architecture for Generalized Expertise).

[14]  “Toward the Construction of a Virtual Yeast.” Nature (2026).

Liujia Qian, Zizhuo Zhou, Peijie Zhou, Bowen Zhou^† , Tiannan Guo, et al.

• AI for Science A virtual cell model for yeast systems biology, advancing AI-driven biological simulation.

[15]  “Intern-S Series: Scientific Multimodal Foundation Models.” Open-source models, 2025–2026.

Shanghai AI Lab Intern-S Team (led by Bowen Zhou^† ).

• AI for Science The Intern-S family (S1, S1-Pro, S2) of scientific multimodal models unified by the SAGE architecture. Achieved Olympiad gold-medal-level mathematical reasoning and top-tier performance across 100+ scientific tasks spanning chemistry, materials, life science, earth science, and physics.

For more publications, please visit:

https://scholar.google.com/citations?hl=fr&tzom=-480&user=h3Nsz6YAAAAJ