Guan-Horng Liu

Machine Learning PhD @ Georgia Tech 🚀

Hi, I am Guan-Horng Liu (I go by "Guan"), a final-year Machine Learning PhD in Georgia Tech.

I study fundamental algorithms for learning diffusion models with optimality structures. I’m actively contributing to nonlinear diffusion models—mainly Schrödinger Bridge and Mirror Diffusion— and large-scale methods for applications in generative modeling, image restoration, unpaired image translation, watermarked generation, opinion depolarization, and single-cell RNA sequencing. Prior to this, I worked on robust architecture-aware neural optimizers.

I’m generally interested in integrating optimality/domain structures into diffusion and flow models, with the goals of enhancing theoretical understanding and developing large-scale algorithms for novel applications. In terms of fundamental research, I combine dynamic optimal transport, stochastic optimal control, and statistical physics. I enjoy applying these tools to a variety of scientific and machine learning problems.

I’m extremely fortunate to intern in FAIR Lab, Meta and Nvidia Research during 2023 and 2022 Summer, working with many talented researchers, including (FAIR) Ricky T. Q. Chen, Yaron Lipman, Maximilian Nickel, Brian Karrer, (Nvidia) Weili Nie, Arash Vahdat, Anima Anandkumar, De-An Huang, (Google DeepMind) Valentin De Bortoli, and (Georgia Tech) Molei Tao. In Georgia Tech, I am advised by Evangelos Theodorou.

See my full CV here (updated in Fed 2024).

Contact: ghliu [at] gatech [dot] edu
Follow: Google Scholar | LinkedIn | ghliu | @guanhorng_liu

Updates
[02/2024] Generalized Schrödinger Bridge Matching (with FAIR, Meta AI) accepted to ICLR.
[09/2023] Two papers, Mirror Diffusion and Momentum Schrödinger Bridge, accepted to NeurIPS.
[05/2023] I have joined FAIR Lab, Meta AI as a research scientist intern this summer.
[04/2023] I co-organize ICML workshop on New Frontiers in Learning, Control, and Dynamical Systems.
[03/2023] Image-to-Image Schrödinger Bridge (with Nvidia Research) accepted to ICML.
[09/2022] Deep Generalized Schrödinger Bridge accepted to NeurIPS (Oral 1.9%).
[02/2022] Likelihood training of Schrödinger Bridge accepted to ICLR.
[09/2021] Three papers on architecture-aware neural optimizers accepted to NeurIPS (Spotlight 3.0%), ICML (Oral 3.0%), and ICLR (Spotlight 3.8%).
Selected Publications
(* Equal contribution, † Equal advising. See Google Scholar for the full list.)
  1. Generalized Schrödinger Bridge Matching
    Guan-Horng Liu, Yaron Lipman, Maximilian Nickel, Brian Karrer, Evangelos A Theodorou, Ricky T. Q. Chen
    International Conference on Learning Representations (ICLR), 2024
    arxiv
  2. Mirror Diffusion Models for Constrained and Watermarked Generation
    Guan-Horng Liu, Tianrong Chen, Evangelos Theodorou†, Molei Tao†
    Advances in Neural Information Processing Systems (NeurIPS), 2023
    arxiv | code
  3. Deep Momentum Multi-Marginal Schrödinger Bridge
    Tianrong Chen, Guan-Horng Liu, Molei Tao, Evangelos Theodorou
    Advances in Neural Information Processing Systems (NeurIPS), 2023
    arxiv | code
  4. I²SB: Image-to-Image Schrödinger Bridge
    Guan-Horng Liu, Arash Vahdat, De-An Huang, Evangelos Theodorou, Weili Nie, Anima Anandkumar
    International Conference on Machine Learning (ICML), 2023
    arxiv | code | project
  5. Deep Generalized Schrödinger Bridge
    Guan-Horng Liu, Tianrong Chen, Oswin So, Evangelos Theodorou
    Advances in Neural Information Processing Systems (NeurIPS), 2022   [Oral 1.9%]
    arxiv | code
  6. Likelihood Training of Schrödinger Bridge using Forward-Backward SDEs Theory
    Tianrong Chen*, Guan-Horng Liu*, Evangelos Theodorou
    International Conference on Learning Representations (ICLR), 2022
    arxiv | code
  7. Second-order Neural ODE Optimizer
    Guan-Horng Liu, Tianrong Chen, Evangelos Theodorou
    Advances in Neural Information Processing Systems (NeurIPS), 2021   [Spotlight 3.0%]
    arxiv | code
  8. Dynamic Game Theoretic Neural Optimizer
    Guan-Horng Liu, Tianrong Chen, Evangelos Theodorou
    International Conference on Machine Learning (ICML), 2021   [Long talk 3.0%]
    arxiv
  9. DDPNOpt: Differential Dynamic Programming Neural Optimizer
    Guan-Horng Liu, Tianrong Chen, Evangelos Theodorou
    International Conference on Learning Representations (ICLR), 2021   [Spotlight 3.8%]
    arxiv