A few constructive approaches to optimal first-order optimization methods in convex optimization

Dec 7, 2021 04:00 PM Singapore (Registration will open at 03:50 PM.)

Join Zoom Meeting:
https://sutd-edu.zoom.us/j/92451603689?pwd=UnpuVWpaOWRwbWpxVmU1MkFZT3BJQT09

Meeting ID: 924 5160 3689
Passcode: kr8.V#bE

Abstract

In this presentation, I want to provide a high-level overview of a recent approach for analyzing and designing first-order methods using symbolic computations and/or semidefinite programming (SDP). This approach is commonly referred to as “performance estimation”. The presentation will be example-based, as the main ingredients necessary for understanding the methodologies are already present in the analysis of base optimization schemes. Relying on those examples, I want to provide an overview on a few principled approaches to the construction of (optimal) first-order methods for convex optimization. The methodology is implemented within two open source packages, allowing to use the framework without the SDP modelling steps.

Papers related to the talk:

  1. Efficient first-order methods for convex minimization: a constructive approach https://link.springer.com/article/10.1007/s10107-019-01410-2
  2. An optimal gradient method for smooth strongly convex minimization https://arxiv.org/pdf/2101.09741.pdf
  3. On the oracle complexity of smooth strongly convex minimization https://www.sciencedirect.com/science/article/pii/S0885064X21000455
  4. Acceleration Methods https://arxiv.org/abs/2101.09545

About the Speaker

Adrien is currently a researcher at INRIA and Ecole Normale Supérieure, in Paris, France. Before that, Adrien obtained a PhD from UCLouvain (2017, Belgium) where he worked under the supervision of François Glineur and Julien Hendrickx. His PhD was followed by a postdoc at INRIA/ENS, where he was supervised by Francis Bach before joining the same team as a researcher. Adrien’s PhD had the chance to be awarded a few local prizes (icteam, IBM-FNRS PhD thesis prizes), a 2017 OPTL best paper award, as well as to be a finalist of the 2015-2018 Tucker prize from the mathematical optimization society.

For more information about the ESD Seminar, please email esd_invite@sutd.edu.sg

Adrien Taylor (INRIA and Ecole Normale Supérieure) - A few constructive approaches to optimal first-order optimization methods in convex optimization