Seminario "Recent perspectives on Frank-Wolfe methods in Machine Learning"

  • Data: 05 febbraio 2015 dalle 11:00 alle 12:00

  • Luogo: Scuola di Ingegneria e Architettura, Aula 5.5, viale del Risorgimento, 2 - Bologna

Il 5 Febbraio 2015 dalle ore 11 alle ore 12 in Aula 5.5 (Viale del Risorgimento 2, Bologna, Aule Nuove) il dr. Emanuele Frandi<>  PhD, post-doc researcher presso ESAT-STADIUS - KU Leuven, Belgio, terrà il seguente seminario:

Title: Recent perspectives on Frank-Wolfe methods in Machine Learning



The Frank-Wolfe algorithm is a classical iterative scheme for convex minimization which has recently been the subject of a renewed interest from researchers. Its solid theoretical properties and sparsity guarantees make it a suitable choice for a wide range of problems arising in Machine Learning, statistics, bioinformatics and other fields. In this seminar, we provide a self-contained introduction to Frank-Wolfe methods and their applications to Machine Learning. We also discuss some variants of the original method and other techniques which can improve on the performance of the standard Frank-Wolfe iteration from both a theoretical and a practical point of view. We motivate our discussion by presenting experimental results on several medium and large-scale SVM datasets.