Short Bio
I am a final year PhD student at Hong Kong University of Science and Technology, supervised by Prof. Jing Tang. My research interests are developing Efficient and Powerful Generative Models, and building Few-Step Text-to-Image/Video Diffusion Models for high-quality and real-time generation.
Academic Service: reviewer for ICML, ICLR, NeurIPS, CVPR, ICCV, etc.
News
- [05/2026] Our TDM-R1 is accepted accpeted to ICML 2026πππ.
- [04/2026] One co-first author paper, DART, is accpeted to SIGIR 2026πππ
- [03/2026] We release TDM-R1, the first RL framework for few-step diffusion models with non-differentiable rewards on scales.
- [01/2026] One first author paper, DGPO, is accpeted to ICLR 2026πππ
- [10/2025] We release DGPO, a novel online RL algorithm for enhacing Diffusion Models. DGPO trains around 20 times faster than prior SOTA and achieves superior performance on both in-domain and out-of-domain reward metrics.
- [09/2025] Two first author papers are accpeted to NeurIPS 2025πππ
- [06/2025] Two first author papers are accpeted to ICCV 2025πππ: 1) TDM β SOTA few-step text-to-image genertion, making few-step student better than its teacher; 2) JDM β the first algorithm that can add additional control unkown to teacher to one-step student.
- [06/2025] We release NCT, a novel approach for adding new controls to one-step generator directly. NCT is the first algorithm that enables adding controls to one-step generator without relying on diffusion distillation or images.
- [04/2025] Delivered an invited talk at ByteDanceΒ about Efficient Post-Training of Diffusion Models (Diffusion Distilattion and RLHF).
- [03/2025] We release R0, a novel conditional generation approach via regularized reward maximization. R0 is the first RLHF algorithm that enables post-training diffusion to few-step text-to-image generator without relying on diffusion distillation or images.
- [03/2025] We release TDM, unifying the trajectory distillation and distribution matching. TDM can distill pre-trained DMs to a few-step generator with better performance with extremely low cost. In particular, TDM distills PixArt to a 4-step generator with 2 A800 hours and outperforms teacher regarding real user preference.
- [03/2025] We release JDM, a method can add additional control unkown to teacher to one-step student. Moreover, JDM decouples the condition learning and fidelity learning, enabling improved usage of classifier-free guidance (CFG) and seamless integration of human feedback learning (HFL).
- [01/2025] Two first/co-first author papers are accepted to ICLR 2025.
Publications
(* denotes co-first authors)
TDM-R1: Reinforcing Few-Step Diffusion Models with Non-Differentiable Reward
Yihong Luo, Tianyang Hu, Weijian Luo, Jing Tang
ICML 2026
[Paper]
Reinforcing Diffusion Models by Direct Group Preference Optimization
Yihong Luo, Tianyang Hu, Jing Tang
ICLR 2026
[Paper]
Noise Consistency Training: A Native Approach for One-Step Generator in Learning Additional Controls
Yihong Luo, Shuchen Xue, Tianyang Hu, Jing Tang
NeurIPS 2025
[Paper]
Reward-Instruct: A Reward-Centric Approach to Fast Photo-Realistic Image Generation
Yihong Luo, Tianyang Hu, Weijian Luo, Kenji Kawaguchi, Jing Tang
NeurIPS 2025
[Paper]
Learning Few-Step Diffusion Models by Trajectory Distribution Matching
Yihong Luo, Tianyang Hu, Jiacheng Sun, Yujun Cai, Jing Tang
ICCV 2025
[Paper]
Adding Additional Control to One-Step Diffusion with Joint Distribution Matching
Yihong Luo, Tianyang Hu, Yifan Song, Jiacheng Sun, Zhenguo Li, Jing Tang
ICCV 2025
[Paper]
You Only Sample Once: Taming One-Step Text-to-Image Synthesis by Self-Cooperative Diffusion GANs
Yihong Luo, Xiaolong Chen, Xinghua Qu, Tianyang Hu, Jing Tang
ICLR 2025
[Paper]
Decoupled Graph Energy-based Model for Node Out-of-Distribution Detection on Heterophilic Graphs
Yuhan Chen*, Yihong Luo*, Yifan Song, Pengwen Dai, Jing Tang, and Xiaochun Cao
ICLR 2025
[Paper]
Energy-Calibrated VAE with Test Time Free Lunch
Yihong Luo, Siya Qiu, Xingjian Tao, Yujun Cai, Jing Tang
ECCV 2024
[Paper]
Fast Graph Sharpness-Aware Minimization for Enhancing and Accelerating Few-Shot Node Classification
Yihong Luo*, Yuhan Chen*, Siya Qiu, Yiwei Wang, Chen Zhang, Yan Zhou, Xiaochun Cao, Jing Tang
NeurIPS 2024
[Paper]
TO-FLOW: Efficient Continuous Normalizing Flows with Temporal Optimization adjoint with Moving Speed
Shian Du*, Yihong Luo*, Wei Chen*, Jian Xu, Delu Zeng
CVPR 2022
[Paper]
LSGNN: Towards General Graph Neural Network in Node Classification by Local Similarity
Yuhan Chen*, Yihong Luo*, Jing Tang, Liang Yang, Siya Qiu, Chuan Wang, Xiaochun Cao
IJCAI 2023
[Paper]
Mitigating Structural Overfitting: A Distribution-Aware Rectification Framework for Missing Feature Imputation
Yifan Song*, Fenglin Yu*, Yihong Luo*, Xingjian Tao, Siya Qiu, Kai Han, Jing Tang
SIGIR 2026
[Paper]
SteadySeg: Improving Maritime Trajectory Staging by Steadiness Recognition
Siya Qiu, Yihong Luo, Qiong Luo, Jing Tang
Ocean Engineering (JCR Q1)
[Paper]
Education
- 2022/02 β 2026/06: Ph.D. in Data Science and Analytics, Hong Kong University of Science and Technology
Experience
- Diffusion RL Research Intern (Qingyun Intern), Tencent HY
- Core contributor to diffusion distillation for Magic Layer, Canva
- Diffusion Post-Training Research Intern, Huawei Noah's Ark Lab
- Research Assistant, Hong Kong University of Science and Technology