Anant Sahai

Timeline

1. 1994–2001

   Degrees at Berkeley and MIT

   He earns a BS in electrical engineering and computer sciences at UC Berkeley in 1994. At MIT, he completes an SM in 1996 and a PhD in 2001, both in electrical engineering and computer science.

2. 2001–2002

   From Enuvis to Berkeley

   In 2001, he works at the wireless startup Enuvis on adaptive signal-processing algorithms for sensitive GPS receivers implemented with software-defined radio. He joins UC Berkeley's Department of Electrical Engineering and Computer Sciences in 2002.

3. 2006

   Anytime capacity and control

   With Sanjoy Mitter, he coauthors a 2006 paper on stabilizing a linear system over a noisy communication link. Their analysis connects channel reliability, expressed through “anytime capacity,” to the ability to keep an unstable system stable.

4. 2012–2017

   Communications and teaching awards

   He receives the IEEE Communications Society's Leonard G. Abraham Prize in 2012. Berkeley gives him its Electrical Engineering Award for Outstanding Teaching in 2017.

5. 2019

   Harmless interpolation of noisy data

   In 2019, he coauthors “Harmless interpolation of noisy data in regression” with Vidya Muthukumar, Kailas Vodrahalli and Vignesh Subramanian. The paper analyzes when linear regression with more features than training examples can fit noisy data and still predict new examples accurately.

6. 2021

   Classification and regression

   The 2021 paper “Classification vs regression in overparameterized regimes: Does the loss function matter?” compares the two tasks in linear models with Gaussian features. He and his coauthors show that sufficient overparameterization can make the minimum-norm least-squares solution identical to a hard-margin support vector machine.

7. 2025–2026

   Mechanisms for in-context recall

   He coauthors “Decomposing Prediction Mechanisms for In-Context Recall,” first posted in 2025 and revised in 2026. Using simple tasks built from dynamical systems, the team identifies separate mechanisms through which transformers recall a sequence and continue predicting its states, with different learning dynamics.

8. As of October 2026

   Professor and faculty chair

   As of October 2026, he is Qualcomm Chair Professor in electrical engineering and computer sciences at UC Berkeley and a part-time Visiting Faculty Researcher at Google. For the 2026–2027 academic year, he chairs the faculty of Berkeley's College of Computing, Data Science, and Society.