cv
Academic background, research experience, publications, and technical skills of Haoran (Haron) Xu.
Education
Sichuan University, College of Biomedical Engineering
B.Eng. candidate in Medical Information Engineering, September 2023 – May 2027
Research Experience
Verifier-Guided Post-training and Inference-time Scaling for RNA Foundation Models
Yale University · Prof. Rex Ying · May 2026 – August 2026
Studied diversity collapse during reward-based post-training and developed KnitNet, a verifier-guided inference-time framework for exact RNA inverse design. Evaluated a 37M-parameter GraphGPS discrete diffusion model on approximately 2,750 RNA structure families and improved exact coverage on OpenKnot57 from 25/57 to 57/57. Optimized the training system to reduce a single B200 run from approximately three days to one day.
Cross-modal Information Routing in Generative Representation Learning
Sichuan University · Prof. Xiao Han · January 2026 – April 2026
Investigated why conditional flow matching can improve deterministic pathology-to-molecular prediction. Used controlled architectural and information-routing interventions across six spatial transcriptomics cohorts and three spatial proteomics cohorts.
BioFlow
Sichuan University · Prof. Xiao Han · September 2025 – December 2025
Developed a support-preserving flow-matching formulation for histology-conditioned spatial transcriptomics prediction. Evaluated the method across eight cancer datasets, obtaining 27–50% improvements in Pearson correlation on major datasets and 5–200× training-efficiency improvements over generative baselines.
SpaMV
Hong Kong Baptist University · Prof. Lu Zhang · July 2024 – May 2025
Contributed to a multi-view variational framework that disentangles shared and modality-specific latent structure in spatial multi-omics. The work was evaluated on simulated and real spatial-omics datasets and published in Nature Communications.
Publications and Manuscripts
- H. Xu and X. Han. “When Does Flow Matching Help Deterministic Multimodal Prediction: A Spatial Transcriptomics Study.” NeurIPS 2026.
- H. Xu, Y. Liu, W. Yuan, and X. Han. “BioFlow: A Biologically Valid Support-Preserving Flow for Histology-Conditioned Spatial Transcriptomics Prediction.” MICCAI 2026.
- Y. Liu, K. Ma, H. Xu, and L. Zhang. “Interpretable spatial multi-omics data integration and dimensionality reduction with SpaMV.” Nature Communications, 2026.
Technical Skills
- Generative modeling: diffusion models, flow matching, discrete diffusion, verifier-guided generation, reinforcement learning, PPO/GRPO, inference-time search
- AI for science: RNA sequence and secondary structure, protein structure, computational pathology, transcriptomics, proteomics, spatial multi-omics
- Research systems: Python, PyTorch, Linux, Git, Slurm/HPC