Michael Mahoney

Also known as Michael W. Mahoney

Michael Mahoney is a researcher whose work links randomized matrix algorithms with scientific machine learning. He earned a Yale PhD in computational statistical mechanics, worked at Yale, Yahoo Research, and Stanford, and joined UC Berkeley Statistics in 2014. At Berkeley, he led the Foundations of Data Analysis Institute; a paper he coauthored on column subset selection and the Nyström method won a 2020 NeurIPS Best Paper Award. At Berkeley Lab, he led a group that developed machine-learning models for earthquake ground motion, then coauthored MatterChat and a study of AI earthquake simulation in 2026. As of October 2026, he is Berkeley Lab's AI Initiative Research Lead, a Professor Adjunct at UC Berkeley, and an ICSI vice president, principal scientist, research mentor, and AI and Big Data group lead.

Now at: University of California, Berkeley, Office of Science and International Computer Science Institute
Works: 510
Citations: 25,769
h-index: 71
Badges: Bridge
Wayfarer

Publication numbers from OpenAlex, as of 22 October 2026.

Timeline

By March 2014

A Yale PhD and research posts

Before joining Berkeley in March 2014, he earns a Yale PhD in computational statistical mechanics and works in Yale's mathematics department, at Yahoo Research, and in Stanford's mathematics department.
March 2014

Joining Berkeley Statistics

In March 2014, he joins UC Berkeley's Statistics Department as Adjunct Associate Professor and works on algorithmic and statistical methods for large-scale data analysis.
October 2017

Leading the FODA Institute

In October 2017, he leads UC Berkeley’s new Foundations of Data Analysis Institute, an NSF-backed project that brings together theoretical statistics, applied mathematics, and theoretical computer science to address foundational data-science questions.
2020

A NeurIPS Best Paper Award

The paper he coauthors with Michal Derezinski and Rajiv Khanna wins a NeurIPS 2020 Best Paper Award for work on column subset selection and the Nyström method. Its analysis derives improved approximation guarantees from matrix spectral properties and establishes a lower bound that shows the multiple-descent curve is inherent to both methods.
2022–2025

Leading Berkeley Lab's AILearn Group

From 2022 until summer 2025, he leads Berkeley Lab's Scientific Data Division AI and Learning Systems Group (AILearn). He develops machine-learning models that use spatiotemporal and domain information to predict earthquake ground motions for seismic hazard assessment. The group studies foundations of scientific machine learning and randomized matrix algorithms.
May 2026

MatterChat for materials science

In May 2026, he coauthors MatterChat, a bridge model that connects a large language model with a physics-based AI for interatomic potentials. The team trains it on nearly 143,000 stable atomic structures and reports better classification of material types and prediction of numerical properties than competing systems.
August 2026

CGM-GM models earthquake motion

In August 2026, he coauthors a study of CGM-GM, a conditional variational autoencoder that generates earthquake ground-motion waveforms. The model simulates thousands of ground motions across the San Francisco Bay Area in minutes on a standard computer with a GPU.
As of October 2026

AI Initiative Research Lead

As of October 2026, he is Berkeley Lab's AI Initiative Research Lead in the AI and Learning Systems Group and Professor Adjunct in UC Berkeley's Statistics Department. At ICSI, he is Vice President, Principal Scientist, Research Mentor, and Group Lead for AI and Big Data.

Career

Since 2022Office of ScienceAI and Learning Systems Group Lead
Since 2014University of California, BerkeleyAdjunct Associate Professor
Since 2013International Computer Science InstituteVice President, Principal Scientist, Research Mentor, AI and Big Data Group Lead
2023–2026Amazon
2007–2014Stanford University
2006–2008Yahoo
1997–2007Yale University