Advances in Empirical Risk Minimization for Image Analysis and Pattern Recognition

Abstract : State of the art approaches in computer vision and medical image analysis are intricately tied to recent advances in machine learning. In order to overcome the inherent variability of visual data, statistical learning techniques are widely applied to extract semantic information from images, and are central to increases in accuracy of systems for image categorization, image retrieval, object detection, and scene analysis. Similarly, analogous learning techniques are beginning to drive advances in areas such as medical image processing, diagnosis, and functional brain image analysis.A core property of learning algorithms is expressed through the no free lunch theorem of machine learning: no given algorithm will have the best possible performance across all problem domains. It is therefore necessary to empirically determine the algorithms that give the best possible performance in specific applications. The problems considered in this manuscript have some commonalities, but also some differences. In computer vision, a common subdomain of problems comes from the desire to perform semantic image analysis, e.g. to specify the key objects in a scene and to possibly infer something about their spatial relationship or functional interactions. In functional brain imaging, we may be interested in identifying regions of the brain implicated in addiction or visual processing and memory. These problems have inherently different goals, but are nevertheless unified by the common goal of analysis of images where we expect spatial regularity. It is in this context that the research described in this manuscript applies, specializes, and extends the state-of-the-art in machine learning for visual data.
Type de document :
HDR
Machine Learning [stat.ML]. ENS Cachan, 2014
Liste complète des métadonnées

Littérature citée [167 références]  Voir  Masquer  Télécharger

https://tel.archives-ouvertes.fr/tel-01086088
Contributeur : Matthew Blaschko <>
Soumis le : lundi 24 novembre 2014 - 15:54:57
Dernière modification le : mardi 5 février 2019 - 13:52:14
Document(s) archivé(s) le : mercredi 25 février 2015 - 10:11:08

Licence


Copyright (Tous droits réservés)

Identifiants

  • HAL Id : tel-01086088, version 1

Collections

Citation

Matthew Blaschko. Advances in Empirical Risk Minimization for Image Analysis and Pattern Recognition. Machine Learning [stat.ML]. ENS Cachan, 2014. 〈tel-01086088〉

Partager

Métriques

Consultations de la notice

694

Téléchargements de fichiers

957