Description
Date depot: 1 janvier 1900
Titre: Reactive Online Deep Learning procedures for streams of data
Directeur de thèse:
Michel CRUCIANU (CEDRIC)
Encadrant :
Marin FERECATU (CEDRIC)
Domaine scientifique: Sciences et technologies de l'information et de la communication
Thématique CNRS : Non defini
Resumé:
We offer a full time PhD position with an expected starting date around March 2016. The duration of the contract is 3 years. This research work will take place both at the CEA LIST - http://www-list.cea.fr/ - in the LADIS department and at the CNAM - http://www.cnam.fr/ - in the team Vertigo - http://cedric.cnam.fr/vertigo/.
Deep learning methods are a set of techniques for learning with neural networks which currently provide the best results to many machine learning problems in image/speech recognition, image scene understanding, natural language processing ... Recent theoretical breakthroughs have render possible the use of deep networks with several hidden layers in a biologically-inspired way [1][2]. Those 'deep' models were until recently considered as difficult to train due to a very high number of parameters [1]. This thesis inscribes itself in the continuation of this breakthrough in machine learning by considering a broad field of applications categorized under the flag 'online learning' on data streams.
Doctorant.e: Besedin Andrey