[Statlist] Research seminar in statistics May 22nd 2015, GSEM University of Geneva

[Eva Cantoni]

Organisers :
E. Cantoni - E. Ronchetti - S. Sperlich - M-P. Victoria-Feser


Friday May 22nd, 2015
at 11h15 - Room M 5220, Uni Mail (40, bd du Pont-d'Arve)


TITLE: SVMs meet metric learning

Alexandros Kalousis
University of Applied Sciences, Western Switzerland


ABSTRACT :
We will discuss different variants of Support Vector Machines (SVMs) 
learned together with linear transformations of the feature spaces on 
which they are applied. Under this scenario the radius of the smallest 
data enclosing sphere is no longer fixed. Therefore optimizing the SVM 
error bound by considering both the radius and the margin has the 
potential to deliver a tighter error bound. We will present a couple of 
algorithms that follow this setting R−SVM$^+_\mu$—a SVM radius-margin 
based feature selection algorithm, and R−SVM$^+$ — a metric 
learning-based SVM. We derive both from a tight approximation of the 
radius and a metric learning interpretation of SVMs. Both optimize 
directly the radius-margin error bound using linear transformations. We 
will present experimental results and compare them to state of the art 
methods such as elastic-net, lasso methods and SVMs. The results 
demonstrate that the incorporation of the radius in the optimization 
problem brings significant advantages. In addition we will examine other 
metric inspired variants of SVMs and their relation to classical metric 
learning algorithms such as LMNN.


Visit the website: http://www.stat-center.unige.ch/ressem.html


-- 
Prof. Eva Cantoni
Research Center for Statistics and
  Geneva School of Economics and Management
University of Geneva, Bd du Pont d'Arve 40, CH-1211 Genève 4
http://www.stat-center.unige.ch/members2/profs/eva-cantoni/