Yang Song
Research Principal at Meta Superintelligence Labs; first author of the score SDE and consistency models papers
- Research Principal, Meta Superintelligence Labs (MSL)1
Yang Song is the Research Principal at Meta Superintelligence Labs (MSL), working with the lab's Chief Scientist to help shape its research direction. Before joining Meta, Song led the strategic explorations team at OpenAI. During a PhD in computer science at Stanford University advised by Stefano Ermon, Song developed core concepts and techniques behind score-based diffusion models, including "Score-Based Generative Modeling through Stochastic Differential Equations", whose official implementation is released under the Apache 2.0 license. Song is also first author of "Consistency Models" and "Improved Techniques for Training Consistency Models", whose code OpenAI publishes under the MIT License, and co-authored work on continuous-time consistency models with Cheng Lu.1234567
Work
Score-based generative modeling through SDEs (external site: github.com)Open-source code
First author, with Jascha Sohl-Dickstein, Diederik P. Kingma, Abhishek Kumar, Stefano Ermon, and Ben Poole, of the 2020 paper; the official implementation in yang-song/score_sde is licensed under the Apache License 2.0.23
Consistency models (external site: github.com)Open-source code
First author of "Consistency Models" (2023, with Prafulla Dhariwal, Mark Chen, and Ilya Sutskever) and "Improved Techniques for Training Consistency Models" (2023, with Prafulla Dhariwal), and co-author with Cheng Lu of "Simplifying, Stabilizing and Scaling Continuous-Time Consistency Models" (2024). OpenAI's consistency_models repository releases code and checkpoints under the MIT License.45671
Other documented roles
Roles the sources document but that are not confirmed as current.
This profile covers public professional work only, from the sources below. It is not an endorsement or a ranking. To correct or remove it, use the correction form.