Loctin, Monsieur Florent (2014) Online Deep Learning for life-long acquisition of visual features PFE - Project Graduation, ENSTA.



This report deals with Deep Learning of visual features in the context of Developmental Robotics. It explores the construction of a deep architecture based on clustering. The goal is to incrementally extract and learn features from images in a fully unsupervised way. The algorithms which are presented, accelerated using parallel computing on GPGPU 1, are able to process a stream of images in real time. Experiments were conduced on the MNIST dataset.

Item Type:Thesis (PFE - Project Graduation)
Uncontrolled Keywords:Robotique développementale, Deep Learning, apprentissage incrémental, clustering, algorithme des k-moyennes, caractéristiques visuelles, champs récepteurs, calcul parallèle
Subjects:Information and Communication Sciences and Technologies
ID Code:6549
Deposited On:24 mars 2015 12:39
Dernière modification:24 mars 2015 12:39

Repository Staff Only: item control page