Premier séminaire de l’axe IA de IBISC, organsé par Blaise HANCZAR et Dominique FOURER.
2 exposés sont proposés:
Orateur: Blaise HANCZAR (PR Univ. Evry, équipe AROBAS)
Titre: Tutorial sur les GAN (Generative Adversarial Network).
Résumé:
Un GAN est un modèle génératif où deux réseaux sont placés en compétition dans un scénario de théorie des jeux. Le premier réseau est le générateur, il génère un échantillon (ex. une image), tandis que son adversaire, le discriminateur essaie de détecter si un échantillon est réel ou bien s’il est le résultat du générateur. Ce tutorial a pour objectif de donner les bases de l’architecture et de l’apprentissage des GAN ainsi que de présenter les dernières avancées dans ce domaine. [Récupérer la présentation au format PDF]
Oratrice: Sarra HOUIDI (post-doctorante Projet ORACLES, équipe SIAM)
Titre: Home Electrical Appliances Recognition using Relevant Features and Deep Neural Networks
Résumé:
In 2018, the residential sector represented 41.8 % of the total energy consumption in France. In this context, information on power consumption per Home Electrical Appliances (HEAs) appears to be an essential requirement for setting up energy management strategies and thus making substantial savings in terms of money and energy. Non-Intrusive Load Monitoring (NILM) methods have emerged as promising solutions to provide a breakdown of the residential energy consumption without instrumenting HEAs. Their identification is made through the analysis of the current and voltage signals using a single measuring device connected to the house electrical panel. There are different ways to address NILM problem and several steps in the NILM pipeline. In this research work we are interested into NILM event-based methods and into the HEAs recognition step through classification methods. We first introduce a new dataset of individual HEAs described by their current and voltage measurements in steady-state conditions. Then, we present the set of physically and meaningful electrical features computed from current and voltage measurements. Since the ultimate goal is to obtain a signature that can uniquely identify single device, selecting the most relevant features is of crucial importance. Therefore, we focus in this research work on the identification of HEAs by investigating Feature Selection (FS) approaches and several classification methods in order to converge to the choice of a suitable association of FS and classification methods that can lead to the best identification performances. The results are compared using our own dataset and a publicily available dataset named PLAID.
- Date: 15/11/2019, 13h30
- Lieu: IBISC, site Pelvoux, Amphithéâtre Bx30
- Organisateur et orateur: Blaise HANCZAR (PR Univ. Evry, IBISC équipe AROBAS)
- Co-organisateur: Dominique FOURER (MCF Univ. Evry, IBISC équipe SIAM)
- Oratrice: Sarra HOUIDI (Post-doctorante projet ORACLES, IBISC équipe SIAM)
