creators

Yang Song: score-based diffusion researcher

Yang Song, research principal at Meta Superintelligence Labs, created score-based generative modeling and works on consistency models for fast sampling.

Yang Song is research principal at Meta Superintelligence Labs, where his site says he works with chief scientist Shengjia Zhao on the lab's research direction. His CV puts him at Meta from 2025; before that he was a member of technical staff at OpenAI from 2022 to 2025, where he led the strategic explorations team. He took a BS in mathematics and physics at Tsinghua University (2012-2016) and a PhD in computer science at Stanford (2016-2022), advised by Stefano Ermon. During the PhD he built up score-based generative modelling. 'Generative Modeling by Estimating Gradients of the Data Distribution' (2019, with Ermon) introduced noise-conditional score networks, and 'Score-Based Generative Modeling through Stochastic Differential Equations' (2021), with five co-authors, unified score-based and diffusion models through stochastic differential equations; it won an ICLR 2021 outstanding paper award. He has since worked on consistency models, which sample in one or two steps: 'Improved Techniques for Training Consistency Models' was an ICLR 2024 oral, and 'Simplifying, Stabilizing and Scaling Continuous-Time Consistency Models', with Cheng Lu, an ICLR 2025 oral.

Specialization

diffusion models, score-based generative modeling, consistency models

Country

United States