Jiahang Xu
Research SDE at Microsoft Research Asia

I am a Research Software Development Engineer at Microsoft Research Asia, focusing on enhancing large language model (LLM) reasoning capabilities. My current work includes test-time scaling method (rStar
), along with ongoing efforts in training-based methods and reinforcement learning from human feedback (RLHF).
Previously, my research centered on efficient and scalable AI systems for hardware-aware neural architectures. I worked on neural architecture search (NAS) with projects like ElasticViT
, SpaceEvo
, and LitePred
, as well as structured pruning techniques such as Compresso
.
Before transitioning to AI systems, my early research during my M.S. at Fudan University focused on medical imaging. I developed foundational frameworks for segmentation, registration, and neurological disease analysis, contributing to early diagnosis of Alzheimer’s and Parkinson’s disease.
selected publications
- ICLR 2024Mutual reasoning makes smaller llms stronger problem-solversarXiv preprint arXiv:2408.06195, 2024
- IEEE Vis 2024VisEval: A benchmark for data visualization in the era of large language modelsIEEE Transactions on Visualization and Computer Graphics, 2024
- ICCV 2023Spaceevo: Hardware-friendly search space design for efficient int8 inferenceIn Proceedings of the IEEE/CVF International Conference on Computer Vision, 2023
- SigKDD 2023Constraint-aware and ranking-distilled token pruning for efficient transformer inferenceIn Proceedings of the 29th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, 2023
- PreprintCompresso: Structured pruning with collaborative prompting learns compact large language modelsarXiv preprint arXiv:2310.05015, 2023
- MedIA 2022Cardiac segmentation on late gadolinium enhancement MRI: a benchmark study from multi-sequence cardiac MR segmentation challengeMedical Image Analysis, 2022
- Front. Neurosci.Computer-Aided Classification Framework of Parkinsonian Disorders Using 11C-CFT PET ImagingFrontiers in aging neuroscience, 2022
- Front. Neurosci.A fully automatic framework for parkinson’s disease diagnosis by multi-modality imagesFrontiers in neuroscience, 2019
- Front. Neurosci.Diagnosis of Alzheimer’s disease via multi-modality 3D convolutional neural networkFrontiers in neuroscience, 2019