Ye Zhu

Princeton University

Position: Postdoctoral Research Associate
Rising Stars year of participation: 2024
Bio

Dr. Ye Zhu is currently a postdoctoral research associate in Computer Science at Princeton University. Her research primarily focuses on multimodal generation, with interdisciplinary work on machine learning for astrophysics. She obtained her Ph.D. in Computer Science from Illinois Institute of Technology, USA, where she received the Award for Excellence in Dissertation in 2023. She earned her M.S. and B.S. degrees from Shanghai Jiao Tong University, China, in 2019 and 2016, respectively. She also holds a French Engineering diploma from Ecole Polytechnique, France. She is a recipient of the ACM Women Scholarship, a regular reviewer for CVPR, ICCV, ECCV, NeurIPS, ICLR, and ICML, and a co-organizer for the Responsible GenerativeAI workshop at CVPR 2024.

Areas of Research
  • Machine Learning
Generative AI: A Sustainable Vision Beyond Scale

Generative AI is transforming society as it progresses from research into the real world with the great success of large-scale generative models in language and vision, such as ChatGPT and StableDiffusion. However, the popularity of these models comes at a significant cost: scaling is highly resource-intensive, requiring vast amounts of data, model parameters, and computational power. My research focuses on making generative models more sustainable, addressing three key aspects: improving generative performance through better multimodal learning frameworks, reducing resource demands by leveraging intrinsic generative dynamics, and extending the impact of generative AI into scientific and social domains through interdisciplinary insights. In the first area, my core research efforts involve creating novel methodological frameworks that enable more precise and versatile applications, such as enhancing the alignment between condition and synthesized data in text-to-image and video-to-music generation. For the second part, my works aim to maintain state-of-the-art performance in a learning-free paradigm via a profound understanding of generative mechanisms from the mathematical and physical perspectives. Lastly, I seek to explore the potential of generative models beyond data synthesis, investigating their ability to learn physical distributions in astrophysics and considering their social impact for real-world deployment.