Yisong Yue (he/him)
California Institute of Technology
1200 E. California Blvd.
CMS, 305-16
Pasadena, CA 91125

Office: 303 Annenberg

Contact Information >

I am a professor of Computing and Mathematical Sciences at the California Institute of Technology. My research interests lie primarily in machine learning, and span the entire theory-to-application spectrum from foundational advances all the way to deployment in real systems. I work closely with domain experts to understand the frontier challenges in applied machine learning, distill those challenges into mathematically precise formulations, and develop novel methods to tackle them.

Asari AI: I am a founding advisor for Asari AI, which aims to develop AI co-inventors (AI agents that can plan, abstract, and verify complex design tasks).

CaineX: I am a founding advisor for CaineX, which aims to develop AI automation for complex back-end medical workflows, notably medical coding.

ICLR Leadership: I previously served as Senior Program Chair of ICLR 2024, and General Chair of ICLR 2025. I am currently a member of the ICLR Board.

My current research interests can be broadly organized into three overlapping groups:

AI for Autonomy: study how AI methods can enable novel capabilities in autonomous systems; characterize and address key technical bottlenecks (e.g., data-driven safety guarantees); deploy in real systems.

Selected Publications

• Morphological-Symmetry-Equivariant Heterogeneous Graph Neural Network for Robotic Dynamics Learning
  Fengze Xie, Sizhe Wei, Yue Song, Yisong Yue, Lu Gan
  Conference on Learning for Dynamics and Control (L4DC), 2025
  [arxiv]
• Robust Agility via Learned Zero Dynamics Policies
  Noel Csomay-Shanklin, William D. Compton, Ivan Dario Jimenez Rodriguez, Eric R. Ambrose, Yisong Yue, Aaron D. Ames
  International Conference on Intelligent Robots and Systems (IROS), 2024
  [arxiv]
• Neural-Fly Enables Rapid Learning for Agile Flight in Strong Winds
  Michael O’Connell*, Guanya Shi*, Xichen Shi, Kamyar Azizzadenesheli, Anima Anandkumar, Yisong Yue, Soon-Jo Chung
  Science Robotics, May 2022.
  [arxiv][online][code][video][press release]
• MLNav: Learning to Safely Navigate on Martian Terrains
  Shreyansh Daftry, Neil Abcouwer, Tyler Del Sesto, Siddarth Venkatraman, Jialin Song, Lucas Igel, Amos Byon, Ugo Rosolia, Yisong Yue, Masahiro Ono
  IEEE Robotics and Automation Letters (RA-L), May 2022
  [conference][journal][video]
• Preference-Based Learning for Exoskeleton Gait Optimization
  Maegan Tucker*, Ellen Novoseller*, Claudia Kann, Yanan Sui, Yisong Yue, Joel Burdick, Aaron D. Ames
  International Conference on Robotics and Automation (ICRA), May 2020.
  (Best Paper Award)
  [pdf][arxiv][demo video][project]

AI for Science: study how AI methods can improve workflows in science and accelerate knowledge discovery; develop methods for automated experiment design and human-intelligible modeling; deploy in real systems.

Selected Publications

• Steering Generative Models with Experimental Data for Protein Fitness Optimization
  Jason Yang, Wenda Chu, Daniel Khalil, Raul Astudillo, Bruce J. Wittmann, Frances H. Arnold, Yisong Yue
  [arxiv][code]
• Population Transformer: Learning Population-level Representations of Neural Activity
  Geeling Chau, Christopher Wang, Sabera Talukder, Vighnesh Subramaniam, Saraswati Soedarmadji, Yisong Yue, Boris Katz, Andrei Barbu
  International Conference on Learning Representations (ICLR), 2025
  (Oral Presentation)
  [project][arxiv]
• InverseBench: Benchmarking Plug-and-Play Diffusion Models for Inverse Problems in Physical Sciences
  Hongkai Zheng, Wenda Chu, Bingliang Zhang, Zihui Wu, Austin Wang, Berthy T. Feng, Caifeng Zou, Yu Sun, Nikola Kovachki, Zachary E. Ross, Katherine L. Bouman, Yisong Yue
  International Conference on Learning Representations (ICLR), 2025
  (Spotlight)
  [project]
• Active Learning-Assisted Directed Evolution
  Jason Yang, Ravi G. Lal, James C. Bowden, Raul Astudillo, Mikhail A. Hameedi, Sukhvinder Kaur, Matthew Hill, Yisong Yue, Frances H. Arnold
  Nature Communications, 2025
  [online][bioRxiv]
• Self-Supervised Keypoint Discovery in Behavioral Videos
  Jennifer J. Sun*, Serim Ryou*, Roni Goldshmid, Brandon Weissbourd, John Dabiri, David J. Anderson, Ann Kennedy, Yisong Yue, Pietro Perona
  IEEE Conference on Computer Vision and Pattern Recognition (CVPR), June 2022.
  [arxiv]
• Task Programming: Learning Data Efficient Behavior Representations
  Jennifer J. Sun, Ann Kennedy, Eric Zhan, David J. Anderson, Yisong Yue, Pietro Perona
  IEEE Conference on Computer Vision and Pattern Recognition (CVPR), June 2021.
  (Best Student Paper Award)
  [arxiv][code][project]

Core AI/ML Research: study the underlying fundamental questions pertaining practical algorithm design, inspired by real-world applications in science and engineering.

Selected Publications

• Find Any Part in 3D
  Ziqi Ma, Yisong Yue, Georgia Gkioxari
  International Conference on Computer Vision (ICCV), 2025
  [project][arxiv][demo][code]
• Practical Bayesian Algorithm Execution via Posterior Sampling
  Chu Xin Cheng, Raul Astudillo, Thomas Desautels, Yisong Yue
  Neural Information Processing Systems (NeurIPS), 2024
  [arxiv]
• ReST-MCTS*: LLM Self-Training via Process Reward Guided Tree Search
  Dan Zhang, Sining Zhoubian, Ziniu Hu, Yisong Yue, Yuxiao Dong, Jie Tang
  Neural Information Processing Systems (NeurIPS), 2024
  [arxiv]
• Online Policy Optimization in Unknown Nonlinear Systems
  Yiheng Lin, James A. Preiss, Fengze Xie, Emile Anand, Soon-Jo Chung, Yisong Yue, Adam Wierman
  Conference on Learning Theory (COLT), 2024
  [arxiv]
• TOTEM: TOkenized Time Series EMbeddings for General Time Series Analysis
  Sabera Talukder, Yisong Yue, Georgia Gkioxari
  Transactions of Machine Learning Research (TMLR), 2024
  [arxiv][code]
• Neurosymbolic Programming
  Swarat Chaudhuri, Kevin Ellis, Oleksandr Polozov, Rishabh Singh, Armando Solar-Lezama, Yisong Yue
  Foundations and Trends in Programming Languages, Volume 7: No. 3, pages 158-243, December 2021.
  [preprint][online]
• Batch Policy Learning under Constraints
  Hoang M. Le, Cameron Voloshin, Yisong Yue
  International Conference on Machine Learning (ICML), June 2019.
  (Oral Presentation)
  [pdf][arxiv][project]

• Formally Verified Code Generation Benchmark -- we present CLEVER, a benchmark for formally verified code generation. Each problem consists of (1) the task of generating a specification that matches a held-out ground-truth specification, and (2) the task of generating a Lean implementation that provably satisfies this specification. [arxiv][code][data]
  
• Steering Generative Models using Experimental Data -- we present a framework for steering generative models using experimental data for protein optimization. [arxiv][code]
  
• Video Inverse Problems -- we developed a diffusion model-based approach for solving video inverse problems. [arxiv][project][code]
  
• Self-Evolving Visual Concept Library -- we developed Escher: Self-Evolving Visual Concept Library using Vision-Language Critic. [project]
  
• InverseBench -- we developed a benchmarking framework for plug-and-play diffusion models for inverse problems in physical sciences. [project]
  
• Active learning-assisted directed evolution -- we developed a new approach for machine learning to accelerate directed evolution for protein optimization. Our method is validated in the wet-lab, where we improved the yield of a desired product of a non-native cyclopropanation reaction from 12% to 93%. [online][bioRxiv]
  
• Find3D -- we developed an approach for training foundation model for 3D part-level understanding. The main contribution is a data engine that can generate large amounts of 3D training data using existing 2D foundation models. [project]
  
• Self-Training for LLM-based Tree-Search -- we developed a self-training approach for process-reward learning to empower LLM-based Tree-Search. [arxiv]
  
• Plug-and-Play Bayesian Inversion via Diffusion Models & Physics -- we developed a principled plug-play inverse imaging approach that blends diffusion model prior and physics-bassed forward models. [arxiv]
  
• Humanlike Bot Behavior -- we developed an approach for generating humanlike bot behavior from demonstrations, demoed in a gym environment within the Fortnite game engine. [project][paper]
  
• Farewell! We had five members depart the group during the 2023-2024 Academic Year:
  • Lu Gan completed her postdoc and has started as a faculty at Georgia Tech.
  • Ziniu Hu completed his postdoc and has started at xAI.
  • Yujia Huang completed her Ph.D. and has started at Citadel Securities.
  • James Preiss completed his postdoc and has started as a faculty at UC Santa Barbara.
  • Kaiyu Yang completed his postdoc and has started at Meta AI.
  
• SceneCraft: An LLM Agent for Synthesizing 3D Scene as Blender Code -- we developed a Large Language Model (LLM) Agent for converting text descriptions into Blender-executable Python scripts which render complex scenes with up to a hundred 3D assets. This process requires complex spatial planning and arrangemet, which we tackle through a combination of advanced abstraction, strategic planning, and library learning. [arxiv]
  
• Tokenized Time Series Embeddings -- we developed a framework for learning a tokenization for downstream time series modeling, including training generalist models that can be applied to many domains. [arxiv][code]
  
• Symbolic Music Generation with Non-Differentiable Rule Guided Diffusion -- we developed a method for guided diffusion using non-differentiable rules, called Stochastic Control Guidance. Our approach is inspired by path integral control and can be applied in a plug-and-play way to any diffusion model. We demonstrate our approach on symbolic music generation. [arxiv][website]