Description
Date depot: 25 octobre 2019
Titre: Neural Architecture Search under budget constraints
Directeur de thèse:
Ludovic DENOYER (Facebook)
Domaine scientifique: Sciences et technologies de l'information et de la communication
Thématique CNRS : Non defini
Resumé:
The objective of the thesis is to develop new neural architecture search methods able to learn the optimal neural network architecture for a particular task under budget constraints (CPU consumption, memory footprint, etc.). The thesis is organized in three steps: use of Reinforcement Learning techniques for discovering static architectures, learning of adaptative architectures, and at last, neural architecture search for lifelong learning.
Doctorant.e: Veniat Tom