PeopleofSuperintelligence

Researcher

Bo Zhao

Bo Zhao is an AI researcher whose work spans dataset condensation, gradient privacy, multimodal language models, and embodied AI. In 2020 and 2021, he coauthored iDLG on gradient leakage and introduced dataset condensation methods using gradient matching, differentiable Siamese augmentation, and distribution matching. His 2022 paper on dataset condensation and privacy won an ICML Outstanding Paper Award. By July 2024, he led the DCAI group as a principal investigator at the Beijing Academy of Artificial Intelligence while coauthoring Bunny, and he went on to coauthor Emu3 in September 2024 and Video-XL at CVPR 2025. A graduate of Peking University and the University of Edinburgh, he is an associate professor in the School of Artificial Intelligence at Shanghai Jiao Tong University and a visiting scholar of Embodied AI at Alibaba Group.

Timeline

  1. January 2020

    Improved Deep Leakage from Gradients

    In January 2020, he coauthors iDLG: Improved Deep Leakage from Gradients with Konda Reddy Mopuri and Hakan Bilen. The paper examines how sharing gradients can leak private training data in distributed learning systems such as collaborative and federated learning.

  2. 2020

    Dataset Condensation with Gradient Matching

    With Konda Reddy Mopuri and Hakan Bilen, he introduces Dataset Condensation with Gradient Matching in 2020, a training set synthesis technique that condenses a large dataset into a small set of informative synthetic samples for training deep neural networks.

  3. 2021

    Differentiable augmentation and distribution matching

    With Hakan Bilen, he coauthors Dataset Condensation with Differentiable Siamese Augmentation at the International Conference on Machine Learning in 2021, condensing large training sets into smaller synthetic sets to train deep neural networks from scratch. In October 2021, he and Bilen also release Dataset Condensation with Distribution Matching to reduce training cost with learned synthetic sets.

  4. 2022

    An ICML Outstanding Paper Award

    In 2022, he coauthors “Privacy for Free: How does Dataset Condensation Help Privacy?” with Tian Dong and Lingjuan Lyu on dataset condensation for private data generation. The paper receives an Outstanding Paper Award at the 39th International Conference on Machine Learning in 2022.

  5. By July 2024

    Principal investigator at BAAI

    By July 2024, he is a principal investigator at the Beijing Academy of Artificial Intelligence and leads its DCAI group. In 2024, he coauthors Bunny, a family of lightweight multimodal language models with flexible vision and language backbones for efficient learning from selected training data.

  6. September 2024

    Emu3 and next-token multimodal models

    In September 2024, he coauthors Emu3, a suite of multimodal models trained solely with next-token prediction. An article on the Emu3 family of multimodal models appears in Nature in January 2026.

  7. 2025

    Video-XL for hour-scale video

    At CVPR 2025, he coauthors Video-XL, an extra-long vision-language model for hour-scale video understanding, as his team has two papers accepted for CVPR 2025 Oral presentations. The model summarizes visual information within each video interval using a Visual Summarization Token.

  8. As of October 2026

    Professor at Shanghai Jiao Tong

    He holds a Ph.D. from the University of Edinburgh and an M.Eng. from Peking University. As of October 2026, he is an associate professor in the School of Artificial Intelligence at Shanghai Jiao Tong University and a visiting scholar of Embodied AI at Alibaba Group. In March 2026, Shuyuan Hu joins his Shanghai Jiao Tong University group to work on real-robot reinforcement learning, and in 2026 Evo-1 receives the Efficient CVPR Badge for computational efficiency.

Career

  1. NowShanghai Jiao Tong UniversityAssociate Professor
  2. NowAlibabaVisiting Scholar of Embodied AI
  3. Beijing Academy of Artificial IntelligencePrincipal Investigator and group lead

Education

  1. University of EdinburghPhD
  2. Peking UniversityM.Eng.

From the biography.

  1. The Massachusetts Institute of Technology trained or employed 44 of the 187 AI major-prize winners, more than any other universityIncludes Bo Zhao. The Massachusetts Institute of Technology trained or employed 44 of the 187 AI major-prize winners, more than any other university. We compare doctorate and 5+ year employment links, career timing and current workplaces for 187 major-prize and 1,111 best-paper winners.
  2. Google alumni have founded or led 133 other organizations, more than any other AI labIncludes Alibaba. We estimate that alumni of 17 major AI labs went on to found or lead 350 other organizations, including 226 they founded, and we trace 472 career moves directly between 23 labs since 2010.
  3. 84% of people shaping AI with recorded undergraduate and doctoral schools changed institutions for their doctorateIncludes Shanghai Jiao Tong University. 84% of people shaping AI with recorded undergraduate and doctoral schools changed institutions for their doctorate. Across 2,116 people with both an undergraduate degree and a doctorate recorded, we trace which undergraduate schools feed which doctoral programs, how often students stay at the same school, and how paths differ by birth-country region.
  4. Universities (31%) and startups and other companies (25%) employ the largest shares of 2020–2026 entrantsIncludes Alibaba. Among 819 people whose first recorded role started in 2020–2026 and who have a known current organization, 60% work in commercial sectors (25% at startups and other companies, 23% at frontier labs and 13% at big tech, with 9 in two), and 31% work at universities, compared with 66% for earlier entrants.

Colleagues

People named in this biography, and people with dated positions at the same organization in the same years.

Sources and credits

Photo: We have no published photo of Bo Zhao yet. Send us a free-licensed photo.

  1. [2001.02610] iDLG: Improved Deep Leakage from Gradients arxiv.org. Checked 22 October 2026.
  2. [2006.05929] Dataset Condensation with Gradient Matching arxiv.org. Checked 22 October 2026.
  3. Dataset Condensation with Differentiable Siamese Augmentation proceedings.mlr.press. Checked 22 October 2026.
  4. [2110.04181] Dataset Condensation with Distribution Matching arxiv.org. Checked 22 October 2026.
  5. [2206.00240] Privacy for Free: How does Dataset Condensation Help Privacy? arxiv.org. Checked 22 October 2026.
  6. Tian Dong's Homepage chichidd.github.io. Checked 22 October 2026.
  7. Home bozhao.me. Checked 22 October 2026.
  8. [2402.11530] Efficient Multimodal Learning from Data-centric Perspective arxiv.org. Checked 22 October 2026.
  9. Modality Alignment for Multimodal Perception sungsoo.github.io. Checked 22 October 2026.
  10. [2409.18869] Emu3: Next-Token Prediction is All You Need arxiv.org. Checked 22 October 2026.
  11. Multimodal learning with next-token prediction for large multimodal models | Nature nature.com. Checked 22 October 2026.
  12. Video-XL: Extra-Long Vision Language Model for Hour-Scale Video Understanding openaccess.thecvf.com. Checked 22 October 2026.
  13. Bo Zhao - 致远学院 en.zhiyuan.sjtu.edu.cn. Checked 22 October 2026.
  14. Shuyuan Hu — Robot Learning shiki42.github.io. Checked 22 October 2026.