Ziqi Ma

About Me

Hello! I am Ziqi, a first-year PhD student at Caltech. Prior to grad school I spent three amazing years at Microsoft where I was exposed to today’s state-of-the-art in ML. Before that I was an undergrad at UChicago double-majoring in CS and statistics. Now I am envisioning what’s next for ML by investigating how it can assist us in solving hard scientific problems that usually require human expertise and creativity.


2023.9.14 - Moved to Pasadena to start my PhD at Caltech as a Kortschak Scholar!
2023.9.5 - Check out this new article I wrote on Data Science@Microsoft medium blog: Many facets of un-truth: LLM hallucination 101. This is a 101 on hallucination that I wish I had when getting started with LLMs. Hope it could be helpful to you too!
2023.4.25 - Check out this new article I wrote on Data Science@Microsoft medium blog: Reasoning about change: Lessons learned from building a near real-time system for Azure pricing. The pricing projects led me to read up on the development of various industrial streaming systems in the past decade. Such a journey from POC to industry standard - quite inspiring!
2023.3.7 - Check out this new article I wrote on Data Science@Microsoft medium blog: Speeding up “Reverse ETL”.
2021.12.21 - Check out this new article I wrote on Data Science@Microsoft medium blog: What do we talk about when we talk about ML robustness. Distribution shift is only part of the story!
2021.8.24 - I got to write about one of my favorite topics - “how/when research becomes useful in practice”, based on interviews with ML practitioners and organization leaders at Microsoft! It’s published on Data Science@Microsoft medium blog: Navigating the (long and winding) road from innovation to production
2021.6.8 - Check out this article I wrote with my co-authors on Data Science@Microsoft medium blog: Data science perspectives on averages and related areas
2021.6.2 - The fair ML compression project I initiated got featured by PureAI: Researchers Explore Intelligent Sampling of Huge ML Datasets to Reduce Costs and Maintain Model Fairness
2020.10.15 - Presented work on circuit learning for quantum sensing at IEEE QCE20 Practical Quantum Sensing Workshop



2020.8-2023.9 (full-time)
2019.6-2019.9 (intern)

Software Engineer II @ AI Platform
AI Platform Responsible AI team: GPT-family model hallucination detection and mitigation
Software Engineer I-II @ Microsoft Cloud Data Sciences
Microsoft Cloud Data Sciences team: full-stack large-scale ML in business domain, including pricing (streaming, patent filed), revenue forecasting, anomaly detection, customer identity
Innovations: initiated collaboration with MSR on loss-based coreset subsampling, prototyped graph model for pricing
Software Engineer Intern @ Azure Intelligence Platform
Reduced support ticket misrouting by building&integrating NLP classifier and implementing distributed log tracing


University of Chicago


B.S. Computer Science& BA Statistics, summa cum laude
GPA: 3.99/4.00
Honors: Enrico Fermi Scholar, Liew Research Scholar, Phi Beta Kappa Society, Dean’s List (all years)


Adaptive Circuit Learning for Quantum Metrology


Ziqi Ma, Pranav Gokhale, Tian-Xing Zheng, Sisi Zhou, Xiaofei Yu, Liang Jiang, Peter Maurer, Frederic T. Chong
In Proceedings of IEEE International Conference on Quantum Computing and Engineering (QCE)
Virtual, Oct 2021


Fair Coreset Construction for Machine Learning

I initiated a collaboration with Microsoft Research on coreset compression to speed up ML training (MSR AI School invited project, 8% acceptance rate), but during implementation realized existing techniques do not consider fairness, which motivated this research as side project.
In the press: Researchers Explore Intelligent Sampling of Huge ML Datasets to Reduce Costs and Maintain Model Fairness

Circuit Learning for Quantum Metrology


I graduated early to lead this collaboration project between members of CS and Molecular Engineering Departments - formulated sensing protocol design as a learning problem by constructing parametrized encoder/decoder circuits; enabled on-device learning from limited, noisy measurements via robust gradient estimation

Thoreau & Water2Cloud


This is an ioT project for environment monitoring, including a soil sensor network deployed in UChicago campus and a water-body sensing system deployed in India. I used machine learning (tree-based models, neural networks, active learning) for water quality (BOD, COD) prediction, and set up the LoRa system (embedded, wireless, backend) for soil sensing.

Learning Trajectory for Everyday Computing

Designed online visual block-based CS-learning games for k-12 education, and analyzed performance of students from Kenwood Academy

Bridges to Work

Built virtual mentor to give disabled students targeted job-seeking advice (rule-based recommendation, running on Django), as member of TechTeam at UChicago’s Institute of Politics

A Little More About Me

I also like to sing, play the piano & pipa (the Chinese lute), and take very long walks.