M. Pawan Kumar

M. Pawan Kumar is an AI researcher whose work spans computer vision, neural-network verification and algorithm discovery. He studied in Hyderabad and Oxford, coauthored OBJ CUT for object-category segmentation, and worked as a postdoctoral researcher at Oxford and Stanford before joining École Centrale Paris's faculty. His later research included branch-and-bound verification of neural networks, error guarantees for physics-informed neural networks, and FunSearch, an evolutionary method for mathematical discovery. In June 2025, he coauthored a white paper presenting AlphaEvolve, an evolutionary coding agent for algorithm discovery; he is an associate professor in Oxford's Department of Engineering Science.

Now at: University of Oxford

Timeline

1999–2003

Undergraduate research in Hyderabad

As an undergraduate at the International Institute of Information Technology, Hyderabad, he works with its Center for Visual Information Technology from 1999 to 2003.
2003–2008

Oxford PhD and OBJ CUT

From 2003 to 2007, he pursues a PhD in Oxford Brookes University's Computer Vision Group. In 2005, he coauthors OBJ CUT with Philip Torr and Andrew Zisserman, a method for automatic object-category segmentation; in 2008, he joins Oxford's Visual Geometry Group as a postdoctoral researcher.
2009–2015

Research at Stanford and Paris

From 2009 to 2011, he works as a postdoctoral researcher with Daphne Koller in Stanford University's Computer Science Department. From 2012 to 2015, he is faculty at École Centrale Paris's Center for Visual Computing.
2020

Branch-and-bound verification

In 2020, he coauthors a branch-and-bound framework for formal verification of piecewise-linear neural networks. It uses a Mixed Integer Linear Programming formulation and a ReLU branching strategy to handle high-input-dimensional problems with convolutional networks.
December 2023

FunSearch and mathematical discovery

In December 2023, he coauthors a paper introducing FunSearch, an evolutionary search method that pairs a pretrained LLM with a systematic evaluator. The method discovers new constructions for the cap set problem.
2024

Error guarantees for PINNs

At ICML in 2024, he coauthors a framework that gives physics-informed neural networks error guarantees across their continuous domain. It introduces ∂-CROWN to bound PINN residual errors.
June 2025

The AlphaEvolve white paper

In June 2025, he coauthors an AlphaEvolve white paper. It presents an evolutionary coding agent powered by large language models for algorithm discovery and optimization.
As of October 2026

Associate professor at Oxford

As of October 2026, he is an associate professor in the Department of Engineering Science at the University of Oxford.