In the paper we propose a Newton approach for the solution of singly linearly-constrained problems subject to simple bounds. Such problems are widespread in real world applications and often have a very large number of variables, like for instance in Support Vector Machine training and Standard Quadratic Programming problems. To tackle this problem we propose a strategy based on the use of a box-conxtrained truncated Newton algorithm on suitable subspaces. We also report a preliminary numerical experience on some SVM training problems to show viability of the proposed approach.
In the paper we propose a Newton approach for the solution of singly linearly-constrained problems subject to simple bounds. Such problems are widespread in real world applications and often have a very large number of variables, like for instance in Support Vector Machine training and Standard Quadratic Programming problems. To tackle this problem we propose a strategy based on the use of a box-conxtrained truncated Newton algorithm on suitable subspaces. We also report a preliminary numerical experience on some SVM training problems to show viability of the proposed approach.
A Truncated Newton Method For Singly Linearly-Constrained Problems Subject To Simple Bounds / Liuzzi, Giampaolo; Lucidi, Stefano; Manno, Andrea; Francesco, Rinaldi. - STAMPA. - (2012). (Intervento presentato al convegno AIRO 2012 tenutosi a Vietri sul Mare nel 4-7/9/2012).
A Truncated Newton Method For Singly Linearly-Constrained Problems Subject To Simple Bounds
Giampaolo Liuzzi;LUCIDI, Stefano;MANNO, ANDREA;
2012
Abstract
In the paper we propose a Newton approach for the solution of singly linearly-constrained problems subject to simple bounds. Such problems are widespread in real world applications and often have a very large number of variables, like for instance in Support Vector Machine training and Standard Quadratic Programming problems. To tackle this problem we propose a strategy based on the use of a box-conxtrained truncated Newton algorithm on suitable subspaces. We also report a preliminary numerical experience on some SVM training problems to show viability of the proposed approach.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.