Researcher

Bowen Zhang

Bowen Zhang is an AI Research Scientist at Meta whose research spans computer vision, multimodal retrieval, and foundation models. He holds a computer-science PhD from USC, BEng and MEng degrees from Tongji, and earlier research roles at Google Brain, Google Research Language, AWS Rekognition, and Tencent AI Lab. His 2018 paper introduced hierarchical embeddings for video and text, and his 2020 work addressed action detection in live streams. He later coauthored STAIR, a sparse CLIP-based retrieval model, and an Apple Intelligence report on on-device and server language models. In July 2025, he left Apple’s foundation-model group for Meta’s superintelligence team.

Now at
Meta
Works
20
Citations
43
h-index
4

Publication numbers from OpenAlex, as of 19 October 2026.

Timeline

  1. 2018

    Hierarchical video-text embeddings

    With Hexiang Hu and Fei Sha, he introduces hierarchical sequence embedding for video and text in 2018, then tests it on retrieval, zero-shot action recognition, and video captioning.

  2. 2020

    Online action detection

    In 2020, he and Hao Chen, Meng Wang, and Yuanjun Xiong design an online action detector for live video streams that uses a buffer to account for broadcast delay.

  3. 2023

    Sparse image-text retrieval

    STAIR extends CLIP into sparse text and image tokens in Zhang’s 2023 paper; the model outperforms CLIP on COCO-5k retrieval and ImageNet tests.

  4. 2024

    Apple foundation models

    In 2024, he coauthors Apple Intelligence Foundation Language Models, a report describing an approximately 3 billion parameter on-device model and a large server model for Private Cloud Compute.

  5. July 2025

    Joining Meta’s superintelligence team

    In July 2025, he leaves Apple’s foundation-model group to join Meta’s superintelligence team.

  6. By 2026

    Research roles at AI labs

    By 2026, his earlier roles include research-intern and student-researcher positions at Google Brain, Google Research Language, AWS Rekognition, and Tencent AI Lab in Seattle.

  7. By 2026

    Degrees from USC and Tongji

    He also holds a computer-science PhD from USC, plus a computer-science BEng and computer-engineering MEng from Tongji University by 2026.

  8. As of October 2026

    AI research scientist at Meta

    As of October 2026, he is an AI Research Scientist at Meta after a staff machine-learning researcher role on Apple AI/ML Foundation Model.

Career

  1. NowMetaAI Research Scientist
  2. Until 2025AppleStaff machine-learning researcher

Education

  1. University of Southern CaliforniaPhD
  2. Tongji UniversityBEng and MEng

Most-cited works

  1. STAIR: Learning Sparse Text and Image Representation in Grounded Tokens2023 · Conference on Empirical Methods in Natural Language Processing (EMNLP) · 17 citations
  2. Apple Intelligence Foundation Language Models2024 · 8 citations
  3. Effect of air distribution mode on jet flame and emission characteristics of high temperature gas-solid mixed fuel2024 · Journal of the Energy Institute · 7 citations
  4. A visible-near-infrared transparent miniaturized frequency-selective metasurface with a microwave transmission window2023 · Nanoscale · 6 citations
  5. Less is More: Removing Text-regions Improves CLIP Training Efficiency and Robustness2023 · 2 citations

Sources and credits

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

  1. [1810.07212] Cross-Modal and Hierarchical Modeling of Video and Text arxiv.org. Checked 18 October 2026.
  2. [2010.03016] Online Action Detection in Streaming Videos with Time Buffers arxiv.org. Checked 18 October 2026.
  3. STAIR: Learning Sparse Text and Image Representation in Grounded Tokens aclanthology.org. Checked 18 October 2026.
  4. [2407.21075] Apple Intelligence Foundation Language Models arxiv.org. Checked 18 October 2026.
  5. Apple pierde a su cuarto investigador de IA en un mes: fue fichado por Meta bloomberglinea.com. Checked 18 October 2026.
  6. Bowen Zhang | Meta zbwglory.github.io. Checked 18 October 2026.