I am currently a senior research scientist within Google Brain. I did my PhD at MIT, working with Suvrit Sra and as a member of the Machine Learning and Learning and Intelligent Systems groups. My research focuses on identifying precise mathematical definitions of diversity to understand the behavior of ML models, e.g., under distribution shift. I wrote my PhD on negatively dependent measures, which use Strongly Rayleigh polynomials to encode desirable properties for diversity modeling.

MITの大学院生の時に、Suvrit Sraと一緒に研究していた。機械学習の数学的な分析と非凸最適化を研究している。その他にも、理論計算機科学や位相幾何学や言語学に興味がある。

[out of date] 我是一个麻省理工学院的研究生。博士導師是Suvrit Sra。研究的课题是人工智能的数学分析和非凸優化。我也对理论计算机科学,拓扑学和语言学感兴趣。



My research focuses on analyzing strongly Rayleigh (SR) measures as a tool to approach machine-learning problems. These measures, among which Determinantal Point Processes have already proven to be of significant interest to the ML community, encode negative dependence between items in subsets of a ground set: they endow the space with repulsive forces between similar points, enabling a careful balancing of the quality and diversity of a subset. I aim to both develop scalable learning and sampling for SR measures over large datasets, and to leverage their properties to guide machine-learning design and analysis.

Recent news

  • [Fall 2019] Paper accepted at NeurIPS 2019: DppNet: Approximating Determinantal Point Processes with Deep Networks
  • [Summer 2019] I joined Google Brain (Cambridge, MA) as a Research Scientist.
  • [April 24th, 2019] I successfully defended my thesis! Slides available here.
  • [Spring 2019] One paper accepted to ICML 2019: Sublinear Sampling for Determinantal Point Processes ((Jennifer Gillenwater, Alex Kulesza, ZM, Sergei Vassilvtiskii)
  • [Winter 2018] Two papers accepted at AISTATS: Learning Determinantal Point Processes by Corrective Negative Sampling (ZM, Mike Gartrell, Suvrit Sra) and Foundations of Sequence-to-Sequence Modeling for Time Series (ZM, Vitaly Kuznetsov)
  • [October 11th] I am giving a tutorial on time series forecasting at the Artificial Intelligence Conference in London
  • [Fall 2018] Two papers accepted at NIPS: Exponentiated Strongly Rayleigh Distributions (ZM, Suvrit Sra, Stefanie Jegelka) and  Maximizing Induced Cardinality Under a Determinantal Point Process (Jennifer Gillenwater, Alex Kulesza, ZM, Sergei Vassilvtiskii)
  • [Fall 2018] I’m continuing as a part-time student researcher at Google Brain in parallel with my PhD program at MIT.
  • [Summer 2018] I am interning at Google Brain (Cambridge) with Jasper Snoek
  • [Spring 2018] I am honored to be a recipient of the 2018 Google PhD Fellowship in Machine Learning


Time series analysis

Time series analysis



Image courtesy of the U.S. Geological Survey

Determinantal Point Processes

Determinantal Point Processes

Elegant and tractable strongly Rayleigh measures, with direct applications to recommender systems and summarization.

Image courtesy of CERN

Strongly Rayleigh measures

Strongly Rayleigh measures

A class of probability measures over subsets of a ground set that enable negative dependence: similar items are pushed away from each other.

Image courtesy of NASA