Mark Peletier

Mark Peletier is a Professor at the Department of Mathematics and Computer Science of the Eindhoven University of Technology (TU/e) in the Netherlands. His research centres on the rigorous mathematical analysis of partial-differential equations, variational problems and gradient flows, recently with applications to machine-learning optimisation. He is a regular speaker at international workshops and conferences.

Research interests

Professor Peletier’s current work focuses on:

• Gradient-descent dynamics and their singular-limit behaviour
• Noise-injection techniques in deep-learning training
• Sharp versus flat minima in loss landscapes
• Variational methods and optimal transport

Recent work

In March 2024 he presented the talk “Singular-limit Analysis of Training with Noise Injection” at the Humboldt University of Berlin. The talk provides a rigorous mathematical framework for understanding how injected noise affects the convergence of gradient descent, and how it biases the solution toward flat minima that generalise better.