Next Thursday September 25, 2025 our team member Martín Hernández will present his PhD Thesis on:
“From Optimal Control to Random and Neural Network Approximation”
Abstract. This talk presents contributions grounded in control and computation. First, we focus on long-time optimal control, particularly the so-called turnpike property—a principle that allows long-horizon problems to behave essentially as steady for most of the time. We analyze its behavior in parameter-dependent settings to show uniformity and applications to singular limits. Second, Random Batch Methods (RBM) approximate time-dependent dynamics by sampling small subproblems over time, yielding computational savings. We exploit these techniques for PDEs (particularly on graph domains) and for nonlinear ODE systems, with applications to Neural ODEs. Finally, we study the expressive power of multilayer perceptrons, showing finite-sample memorization (simultaneous controllability) and universal approximation in a narrow-but-deep regime.
WHEN
Thu. September 25, 2025 at 12:30H
WHERE
WiSo Erweiterungsbau, SR 429, 2. OG
Lange Gasse 20, Nürnberg
Board of examiners
• Prof. Dr. Wiedemann
• Prof. Dr. Zuazua
• Prof. Dr. Bänsch
|| You might like: Upcoming events
