Within a statistical learning setting, we propose and study an iterative regularization algorithm for least squares defined by an incremental gradient method. In particular, we show that, if all other parameters are fixed a priori, the number of passes over the data (epochs) acts as a regularization parameter, and prove strong universal consistency, i.e. almost sure convergence of the risk, as well as sharp finite sample bounds for the iterates. Our results are a step towards understanding the effect of multiple epochs in stochastic gradient techniques in machine learning and rely on integrating statistical and optimization results

Learning with incremental iterative regularization

Lorenzo Rosasco;Silvia Villa
2015-01-01

Abstract

Within a statistical learning setting, we propose and study an iterative regularization algorithm for least squares defined by an incremental gradient method. In particular, we show that, if all other parameters are fixed a priori, the number of passes over the data (epochs) acts as a regularization parameter, and prove strong universal consistency, i.e. almost sure convergence of the risk, as well as sharp finite sample bounds for the iterates. Our results are a step towards understanding the effect of multiple epochs in stochastic gradient techniques in machine learning and rely on integrating statistical and optimization results
File in questo prodotto:
File Dimensione Formato  
rosasco11567-888561.pdf

accesso aperto

Descrizione: Contributo principale
Tipologia: Documento in Post-print
Dimensione 570.38 kB
Formato Adobe PDF
570.38 kB Adobe PDF Visualizza/Apri
nips2015_suppl_final.pdf

accesso aperto

Tipologia: Altro materiale allegato
Dimensione 411.57 kB
Formato Adobe PDF
411.57 kB Adobe PDF Visualizza/Apri

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/11567/888561
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus 46
  • ???jsp.display-item.citation.isi??? 0
social impact