Yang Song

Yang Song is an incoming Assistant Professor in Electrical Engineering and Computing and Mathematical Sciences at the California Institute of Technology (Caltech). Prior to joining Caltech, he leads the Strategic Explorations team at OpenAI. He received his Ph.D. in Computer Science from Stanford University under the supervision of Stefano Ermon and completed his Bachelor's degree in Mathematics and Physics from Tsinghua University.

His educational background includes:

• Ph.D. in Computer Science, Stanford University
• Bachelor's in Mathematics and Physics, Tsinghua University

Dr. Song's research focuses on building powerful AI models capable of understanding, generating, and reasoning with high-dimensional data across diverse modalities. He is particularly known for inventing foundational concepts and techniques in score-based diffusion models, which have revolutionized the field of generative AI. His work bridges theoretical advances with practical applications, particularly in image generation, medical imaging, and solving inverse problems. His research has demonstrated how score-based models can achieve state-of-the-art results in image generation while maintaining flexibility for various applications including medical image reconstruction.

Analysis of his publication record reveals a strong trajectory in generative modeling, with a particular emphasis on score-based approaches and diffusion models. His work consistently addresses fundamental challenges in generative modeling including sample quality, training stability, computational efficiency, and application to real-world problems. His most recent work on consistency models represents a significant advancement toward making generative models practical for real-time applications.

His notable achievements include:

• ICLR 2021 Outstanding Paper Award for Score-Based Generative Modeling through Stochastic Differential Equations
• NeurIPS 2021 Spotlight Presentation for Maximum Likelihood Training of Score-Based Diffusion Models
• Multiple ICLR Oral presentations for his work on consistency models
• Developing foundational techniques that power many modern AI image generation systems

His GitHub repository for score-based generative modeling has gained significant traction in the research community, with over 1,700 stars, reflecting the impact of his work. His research bridges theoretical machine learning with practical applications, particularly in medical imaging where his techniques have shown promise for improving image reconstruction in CT and MRI.