Sébastien Colla¶
News¶
- (February 2023) We have organized the first PEP-talks workshop in Louvain-la-Neuve. This workshop is focused on the theory and applications of Performance Estimation Problems (PEP) in continuous optimization.
- (December 2022) I was at IEEE 61th CDC conference to present one of our last results
- Sebastien Colla, Julien M. Hendrickx, “Automated Performance Estimation for Decentralized Optimization via Network Size Independent Problems”, in 61th IEEE Conference on Decision and Control (CDC), 2022. [PDF, arXiv, slides,]
Research Interests¶
- Decentralized Optimization
- Computer-aided worst-case analyses
- Multi-Agent system.
The goal of my thesis is to boost the research in decentralized optimization by creating tools allowing one to automatically compute the worst-case performance of any decentralized optimization algorithm, and to identify the bottleneck instances, providing insight on what limits the performance. This will contribute to a better understanding of decentralized optimization algorithms and enable rapid exploration of new algorithms ideas and their iterative improvement.
This project relies on the Performance Estimation Problem (PEP) approach developed by Adrien Taylor for classical centralized optimization during his thesis. It relies on formulating the evaluation of an algorithm worst-case performance as an optimization problem itself, whose variables are the objective functions and initial conditions. Such problems are very complex, but have been shown to be solvable exactly. The PESTO toolbox allows to write and solve them easily with Matlab. The equivalent Python toolbox, called PEPit, has been released in 2022.
Here is a Vulgarized Summary of my research project written in March 2021.