Yu Xie

Also known as 谢玙

Yu Xie is a Senior Researcher in Microsoft Research’s AI for Science group in Berlin, where she works on AI for computational molecular and material sciences. At Harvard, she earned a PhD with Boris Kozinsky and developed Bayesian force fields for stanene, catalytic hydrogen and silicon carbide. She later coauthored BioEmu-1, which Microsoft released as open source, and a Science paper that emulated protein ensembles by generating thousands of statistically independent structures per hour on a single GPU. Her 2026 preprint introduced diffusion sampling for rare molecular events while preserving unbiased thermodynamic estimates.

Now at: Microsoft Research

Timeline

2020

Mapped force fields for stanene

In August 2020, she coauthors a preprint with Jonathan Vandermause, Lixin Sun, Andrea Cepellotti and Boris Kozinsky that speeds Gaussian-process force fields by mapping forces and uncertainties onto low-dimensional features. The method simulates the stability of a stanene monolayer and reveals an unusual phase transformation in 2D stanene.
September 2022

Reactive force fields for H/Pt

Nature Communications publishes her paper with Jonathan Vandermause, Jin Soo Lim, Cameron J. Owen and Boris Kozinsky in September 2022, applying Bayesian active learning to hydrogen turnover on a Pt(111) catalyst surface. At Harvard’s John A. Paulson School of Engineering and Applied Sciences, she implements the on-the-fly training workflow coupled with ASE.
March 2023

Silicon carbide polymorphs in days

In March 2023, her first-author paper with Jonathan Vandermause, Senja Ramakers, Nakib H. Protik, Anders Johansson and Boris Kozinsky builds a sparse Gaussian-process force field on atomic cluster expansion descriptors. The workflow trains silicon-carbide polymorph models in a few days of computer time.
October 2024

A Harvard PhD and Berlin colloquium

In October 2024, she presents her silicon-carbide work at a Berlin colloquium while working as a senior researcher at Microsoft AI for Science in Berlin. Her Harvard PhD work with Boris Kozinsky focuses on machine-learning force fields and Bayesian active learning for molecular-dynamics simulations of material phase transitions.
February 2025

Open-sourcing BioEmu-1

Microsoft Research releases BioEmu-1 as open source in February 2025. The deep-learning model gives scientists a view of the structural ensembles each protein can adopt, and Xie is among its authors.
July 2025

BioEmu in Science

Science publishes “Scalable emulation of protein equilibrium ensembles with generative deep learning” in July 2025, with Xie among its authors. BioEmu emulates protein equilibrium ensembles by generating thousands of statistically independent structures per hour on a single GPU.
February 2026

Diffusion models for rare events

In February 2026, she coauthors a preprint that introduces enhanced diffusion sampling to explore rare-event regions while preserving unbiased thermodynamic estimates. The methods estimate equilibrium properties across toy systems, protein-folding landscapes and folding free energies in GPU-minutes to hours per system.
As of October 2026

Senior researcher in AI for Science

As of October 2026, she is a Senior Researcher in Microsoft Research’s AI for Science group in Berlin, working on artificial intelligence for computational molecular and material sciences. In July 2026, she speaks in the Meet the AI for Science Team Q&A at Microsoft Research’s ICML booth.