Ashia Wilson

Timeline

1. 2011

   Applied mathematics at Harvard

   She earns an AB in applied mathematics from Harvard University in 2011.

2. 2016

   A framework for accelerated optimization

   In 2016, she coauthors A Variational Perspective on Accelerated Methods in Optimization with Andre Wibisono and Michael I. Jordan. The paper introduces the Bregman Lagrangian, a mathematical framework that generates a broad class of accelerated optimization methods in continuous time.

3. 2017

   Testing adaptive gradient methods

   With Rebecca Roelofs, Mitchell Stern, Nati Srebro, and Benjamin Recht, she coauthors The Marginal Value of Adaptive Gradient Methods in Machine Learning in 2017. Their experiments find that adaptive methods can produce models that perform less well on new data than stochastic gradient descent.

4. 2018

   A statistics PhD at Berkeley

   She completes a statistics PhD at the University of California, Berkeley in 2018, advised by Michael Jordan and Ben Recht. Her dissertation is Lyapunov Arguments in Optimization.

5. 2020

   A Microsoft Research postdoc

   In 2020, she works as a postdoctoral researcher in the Machine Learning Group at Microsoft Research, New England. Her research examines guarantees for algorithm performance and frameworks for fairness and governance in machine learning.

6. January 2021

   Joining the MIT faculty

   In January 2021, she joins MIT's Department of Electrical Engineering and Computer Science as an assistant professor. She joins the Faculty of Artificial Intelligence and Decision-Making and the Laboratory for Information and Decision Systems in 2021.

7. 2024

   Algorithmic Pluralism wins at FAccT

   In 2024, Algorithmic Pluralism: A Structural Approach To Equal Opportunity wins a Best Paper Award at the ACM Conference on Fairness, Accountability, and Transparency. She coauthors the paper with Shomik Jain, Vinith Suriyakumar, and Kathleen Creel. The paper proposes reducing barriers to opportunity across systems of algorithmic decisions.

8. September 2025

   A framework for AI benchmarks

   In September 2025, she and Nathanael Jo release What Does Your Benchmark Really Measure? A Framework for Robust Inference of AI Capabilities. They propose treating AI evaluation as inference from a theory of capability and develop methods that account for uncertainty from test variations and finite samples.

9. 2026

   MIT's Junior Bose Award

   She receives MIT's Junior Bose Award in 2026 for her contributions to education.

10. As of October 2026

    Leading MIT's OSE Lab

    As of October 2026, she is the Lister Brothers Career Development Associate Professor in MIT's Department of Electrical Engineering and Computer Science and directs the Optimization, Safety, and Evaluation Lab. Her group studies efficient optimization and sampling, AI safety, and methods for evaluating AI capabilities and social effects.