Preprints & Publications

(In reversed chronological order. * denotes equal contribution or alphabetical order.)

2026

  1. rethink.jpg
    ICML 2026
    Rethinking the Design Space of Reinforcement Learning for Diffusion Models: On the Importance of Likelihood Estimation Beyond Loss Design
    Jaemoo Choi*Yuchen Zhu*Wei GuoPetr MolodykBo Yuan, Jinbin Bai, Yi Xin, Molei Tao, and Yongxin Chen
    In Forty-third International Conference on Machine Learning, 2026
    TL;DRShows that ELBO-based likelihood estimation is a simple yet effective design axis for RL with diffusion models.
  2. metadns.jpg
    ICML 2026
    MetaDNS: Enhancing Exploration in Discrete Neural Samplers via Metadynamics
    Xiaochen DuJuno NamJaemoo ChoiWei Guo, Sathya Edamadaka, Junyi Sha, Elton Pan, Yongxin ChenMolei Tao, and Rafael Gómez-Bombarelli
    In Forty-third International Conference on Machine Learning, 2026
    TL;DRProposes a metadynamics-inspired technique to enhance exploration in discrete neural samplers.
  3. jeais.jpg
    ICLR 2026
    Complexity Analysis of Normalizing Constant Estimation: from Jarzynski Equality to Annealed Importance Sampling and beyond
    Wei GuoMolei Tao, and Yongxin Chen
    In The Fourteenth International Conference on Learning Representations, 2026
    TL;DRGives a unified complexity analysis for normalizing constant estimation methods: from Jarzynski equality to annealed importance sampling and reverse diffusion sampler.
  4. dasbs.jpg
    ICML 2026
    Discrete Adjoint Schrödinger Bridge Sampler
    In Forty-third International Conference on Machine Learning, 2026
    TL;DRAn authentic extension of adjoint matching to discrete state spaces: adjoint matching = target matching + fixed-point iteration.
  5. pdns.jpg
    ICLR 2026
    Proximal Diffusion Neural Sampler
    In The Fourteenth International Conference on Learning Representations, 2026
    TL;DRProximal point updates on path-measure space for stable and efficient neural sampler training.
  6. replaid.jpg
    Preprint
    Continuous Diffusion Scales Competitively with Discrete Diffusion for Language
    arXiv preprint arXiv:2605.18530, 2026
    TL;DRWe establish the first scaling law for continuous diffusion language models (DLMs) that rivals discrete DLMs.
  7. dmpo.jpg
    ICML 2026 Spotlight
    Enhancing Reasoning for Diffusion LLMs via Distribution Matching Policy Optimization
    In Forty-third International Conference on Machine Learning (Spotlight, top 2.2%), 2026
    TL;DRDistribution matching policy optimization via weighted denoising cross-entropy: a new RL paradigm beyond policy gradients.

2025

  1. almc.jpg
    ICLR 2025
    Provable Benefit of Annealed Langevin Monte Carlo for Non-log-concave Sampling
    Wei GuoMolei Tao, and Yongxin Chen
    In The Thirteenth International Conference on Learning Representations, 2025
    TL;DRProvides a theoretical guarantee of the convergence of annealed Langevin Monte Carlo from the perspective of optimal transport and Girsanov’s theorem.
  2. fast_solver.jpg
    NeurIPS 2025
    Fast solvers for discrete diffusion models: Theory and applications of high-order algorithms
    In The Thirty-ninth Annual Conference on Neural Information Processing Systems, 2025
    TL;DRDevelops high-order numerical solvers that accelerate discrete diffusion models with theoretical guarantees and practical gains.
  3. mdns.jpg
    NeurIPS 2025
    MDNS: Masked Diffusion Neural Sampler via Stochastic Optimal Control
    In The Thirty-ninth Annual Conference on Neural Information Processing Systems, 2025
    TL;DRFrames masked diffusion neural sampling as a stochastic optimal control problem for efficient discrete neural sampler training.

2024

  1. Preprint
    Plug-and-Play Controllable Generation for Discrete Masked Models
    Wei Guo*Yuchen Zhu*Molei Tao, and Yongxin Chen
    arXiv preprint arXiv:2410.02143, 2024
    TL;DRDevelops a plug-and-play control method for steering masked discrete diffusion models.

2023

  1. Undergrad Thesis
    Theoretical Analysis of the Approximation Properties of Score-Based Generative Models
    Wei Guo
    Undergraduate Thesis, School of Mathematical Sciences, Peking University, 2023
    TL;DRStudies the convergence guarantees of score-based generative models given an imperfect score estimator and discretization errors.