Jennifer J. Sun

Timeline

1. 2020–2021

   Pose embeddings at ECCV

   After a bachelor's degree in engineering science (electrical and computer engineering) at the University of Toronto, she pursues a Caltech PhD in computing and mathematical sciences. In 2020, she and her coauthors introduce Pr-VIPE, a spotlight paper at ECCV 2020 that maps 2D body pose keypoints to a view-invariant embedding space. She writes the Google Research blog post on it in January 2021 as a student researcher.

2. 2021

   Best student paper at CVPR

   She wins Best Student Paper at CVPR 2021 for Task Programming: Learning Data Efficient Behavior Representations, which studies how to use programmatic weak supervision to reduce the need for supervised labels. Her Caltech PhD is co-advised by Pietro Perona and Yisong Yue.

3. 2023

   The MABe22 behavior benchmark

   She is first author of MABe22, a multi-agent video and trajectory benchmark for learned behavior representations, published at ICML 2023. The data covers triplets of interacting mice, symbiotic beetle-ant interactions and groups of interacting flies, with coauthors including Pietro Perona, Yisong Yue and Ann Kennedy.

4. 2023–2024

   Research scientist at Google

   By September 2023, she is a research scientist at Google while finishing her Caltech PhD, and she spends a year in the role. In 2024, she is among the authors of VideoPrism, a foundational visual encoder for video understanding that Google Research introduces in February 2024.

5. 2024

   Assistant professor at Cornell

   She joins Cornell's Department of Computer Science as an assistant professor in fall 2024. Her work spans computer vision and machine learning, including representation learning, program synthesis, generative modeling and human-AI collaboration.

6. February 2025

   AI for animal welfare

   In February 2025, she is working with veterinarians at the College of Veterinary Medicine to explore whether large language models and vision models could help analyze animal health records and visual data and support diagnoses in the clinic.

7. June 2025

   Language models learn hidden Markov models

   In June 2025, she coauthors a paper with Yijia Dai, Zhaolin Gao, Yahya Sattar and Sarah Dean showing that pre-trained large language models can model data generated by hidden Markov models through in-context learning. On real-world animal decision-making tasks, the approach reaches competitive performance with models that human experts designed.

8. As of October 2026

   SunLab at Cornell

   She is an assistant professor of computer science at Cornell, where her group SunLab builds intelligent systems that collaborate with scientists across machine learning, ecology, neuroscience and language modeling. She is also a part-time research scientist at Google DeepMind.