Hi! I’m Sulin. CV
I am a postdoctoral researcher at MIT in Rafael Gómez-Bombarelli ’s group, and I also collaborate with Tommi Jaakkola ’s group. I received my PhD from Princeton University in 2023, advised by Ryan P. Adams and Peter J. Ramadge . In summer 2021, I interned at Meta Research with Ben Letham and Eytan Bakshy in the Adaptive Experimentation team. I did my undergrad at National University of Singapore and worked with Sinno Jialin Pan at Nanyang Technological University, Singapore.
My research interests are broadly in machine learning, with a current focus on generative models:
- Planning in Diffusion/Masked Language Models: [DDPD: Discrete Diffusion with Planned Denoising ] introduces the first use of a planning model for guided self-correction in the denoising process, significantly improves generation quality compared to SOTA on GPT-2 scale language modeling and ImageNet 256 x 256.
- Scalable Training and Inference of Autoregressive Models: [Generative Marginalization Models ] scalable alignment with reward model, over 1000× speedup in marginal inference
- Amortized Inference/Optimization: [Marginal Inference , Amortized Hyperparam. Optimization ]
- Interpretability and Safety in Knowledge Discovery: [SEBO: Sparse Bayesian Optimization , ProBF: Probabilistic Safety Certificates ]
My work centers on innovating in synthetic data generation, designing complex multi-component AI systems, and deriving scalable training objectives grounded in first principles. I am passionate about using generative AI to tackle impactful real-world problems and drive advances in the next generation of AI-powered systems.
Representative Publications
Think While You Generate: Discrete Diffusion with Planned Denoising. Paper
| Code
Sulin Liu, Juno Nam, Andrew Campbell, Hannes Stärk, Yilun Xu, Tommi Jaakkola, Rafael Gómez-Bombarelli.
Under submission, 2024.
Flow Matching for Accelerated Simulation of Atomic Transport in Materials. Paper
Juno Nam, Sulin Liu, Gavin Winter, KyuJung Jun, Soojung Yang, Rafael Gómez-Bombarelli.
Under submission, 2024.
short version at ICML Workshop on Machine Learning for Life and Material Science: From Theory to Industry applications
(Best paper in Material Science track, 2/141)
Generative Marginalization Models. Paper
| Code
| Video
Sulin Liu, Peter J. Ramadge, Ryan P. Adams.
International Conference on Machine Learning (ICML), 2024.
short version at ICML Workshop on Structured Probabilistic
Inference & Generative Modeling. (Contributed talk, 6/125)
Sparse Bayesian Optimization. Paper
| Code
| Tutorial
| Video
Sulin Liu* (equal contr.), Qing Feng*, David Eriksson*, Benjamin Letham, Eytan Bakshy.
International Conference on Artificial Intelligence and Statistics (AISTATS), 2023.
short version at at NeurIPS Workshop on Gaussian Processes, Spatiotemporal Modeling, and Decision-making Systems. (Contributed talk, top 5 selected)
Task-Agnostic Amortized Inference of Gaussian Process Hyperparameters. Paper
| Code
| Slides
| Video
Sulin Liu, Xingyuan Sun, Peter J. Ramadge, Ryan P. Adams.
Advances in Neural Information Processing Systems (NeurIPS), 2020.
short version at 7th ICML Workshop on Automated Machine Learning. (Spotlight talk)
Recent News
- 10/2024: Gave a talk on DDPD at Google DeepMind Seminar Series on Generative Modeling, Sampling and Transport.
- 10/2024: Preprint of DDPD is out on arXiv . Check out the summary of the paper in this thread !
- 05/2024: Generative Marginalization Models accepted to ICML 2024!
- 12/2023: Invited to talk at Hong Kong University Institute of Data Science.
- 09/2023: Started my postdoc at MIT.