| Zeit: | 11. Juni 2024 |
|---|---|
| Download als iCal: |
|
Dr. Luca Furieri
Swiss Federal Institute of Technology in Lausanne
Lausanne, Switzerland
Tuesday 2024-06-11 4 p.m.
IST Seminar Room 2.255 - Pfaffenwaldring 9 - Campus Stuttgart-Vaihingen
Abstract
As we tackle complex optimization problems in engineering and computer science, developing algorithms that converge faster and find better solutions is crucial. While control theory provides optimal worst-case convergence rates for convex functions, a recent trend in machine learning named Learning to Optimize (L2O) uses neural networks to discover update rules that excel in complex scenarios; the catch is, formal convergence guarantees are not available. Our research bridges these two paradigms using nonlinear system theory. We introduce an unconstrained parametrization of all convergent algorithms for smooth, non-convex functions. Our framework is compatible with automatic differentiation tools, ensuring convergence while learning to optimize over complex landscapes.
Biographical Information
Luca Furieri is a Principal Investigator at EPF Lausanne since 2023. His research focuses on
optimal control and optimization for distributed decision-making and large-scale cyber-physical
systems. Previously, he has been a Postdoctoral researcher at the Automatic Control Laboratory,
EPFL. In 2020, he has been awarded a Ph.D. degree in Control and Optimization from ETH - Zurich. He
has received the SNSF Ambizione career grant in 2022, the IEEE Transactions on Control of Network
Systems Best Paper Award in 2022, and the American Control Conference O. Hugo Schuck Best Paper
Award in 2018.