Chang Xu

About

Dr. Chang Xu is a Senior Researcher in the Machine Learning Group at Microsoft Research Asia. She has extensive expertise in advancing fundamental machine learning algorithms, including representation learning, network architecture design, and evolving learning paradigms (such as reinforcement, generative, contrastive, invariant, and causal learning). Her research also covers multi-modal learning across various modalities (text, time series and image data) and Large Language Models (LLMs), focusing on foundation models, reasoning, and agentic systems. She is committed to applying these techniques to real-world scenarios, including AI in finance, healthcare, and storage systems.

Research Interests

Time Series Analysis and Generation

Time-series data is a cornerstone of knowledge and data engineering, capturing the dynamic evolution of real-world systems. Dr. Xu’s research addresses the full spectrum of time-series analysis, from multi-granularity information fusion and forecasting to interpretability and anomaly detection. Currently, she identifies Time Series Generation (TSG) as a transformative frontier in the field. Driven by the demand for reliable data augmentation, controllable simulation, and privacy-preserving data sharing, she leverages Generative AI (particularly Diffusion Models) to reshape the methodological landscape of TSG. Her research explores several cutting-edge dimensions, including controllable and causal generation for task-aware risk evaluation, multimodal integration for semantically-guided synthesis, and continuous generation for evolving systems. Additionally, she investigates how TSG can provide online data augmentation to empower the pre-training and generalization of Time Series Foundation Models (TSFMs). To support the research community, she led the development of TimeCraft, a comprehensive open-source library for advanced, controllable, and cross-domain time-series generation.
Project [TimeCraft]: https://github.com/microsoft/TimeCraft

AI for Healthcare

Dr. Xu is dedicated to transforming healthcare by building diagnostic foundation models and agentic systems that integrate multi-modal clinical data, including EHR, clinical notes, images, and audio. She led the research and development of MIRA, a unified medical time-series foundation model designed to capture universal physiological patterns from real-world clinical data. Pretrained on a massive corpus of over 454 billion time points, MIRA explicitly addresses core clinical challenges such as irregular sampling, heterogeneous frequencies, and pervasive missing values. The model introduces key technical innovations including Continuous-Time Positional Encoding (CT-RoPE) for variable time intervals, a Frequency-Specialized Mixture-of-Experts architecture for handling diverse temporal resolutions, and Neural ODE-based extrapolation for modeling latent physiological dynamics in continuous time.
Project [MIRA]: https://github.com/microsoft/MIRA

AI in Finance

Earlier in her career, Dr. Xu focused on deep quantitative investment and financial market simulation. She developed advanced models for stock price prediction by integrating heterogeneous data sources, ranging from coarse-grained trends (daily or weekly) to fine-grained dynamics at the minute level. Her research also incorporated event-driven forecasting using financial news and high-frequency order book (LOB) data to model market microstructure and short-term movements. Furthermore, she participated in the development of MarS, a large-scale financial market simulation engine powered by the Large Market Model (LMM). MarS enables realistic, interactive, and controllable order generation, serving as a powerful platform for “what-if” analysis and reinforcement learning agent training.
Project [MarS]: https://github.com/microsoft/MarS

Professional Background & Achievements

Prior to joining MSRA, Dr. Xu received her B.S. degree in Computer Science from Nankai University in 2014 and her Ph.D. in 2019 through the Joint PhD Program between Microsoft Research Asia and Nankai University.

She has published dozens of papers in top-tier journals and conferences, including ICLR, ICML, NeurIPS, KDD, WWW, AAAI, IJCAI, EMNLP, MM, CIKM, ICME, IEEE TKDE, and IEEE TOC, with over 1,400 citations. She is an active member of the academic community, organizing workshops at top conferences and serving as a PC member and reviewer for leading venues.

Collaboration & Recruitment

Dr. Xu is actively seeking collaboration opportunities in the areas of Time Series, Generative Modeling, AI for Healthcare, and LLMs. She warmly welcomes scholars and students to reach out for discussions.

Internship Opening: She is also looking for self-motivated interns to work on these exciting directions. If you are interested, please send your CV to chanx@microsoft.com.