Titre : Sequential Decision Modeling in Uncertain Conditions

Jury :

Abstract:

Sequential decision modeling in uncertain conditions encompasses tasks in generative sequence modeling, structured prediction, planning, and reinforcement learning. This talk will define and demonstrate methods to improve these areas, using conditional information and hierarchical structure to learn control dynamics in compressed space. A number of practical systems will be surveyed, and this talk will include results for handwriting synthesis, text to speech synthesis, harmonic recomposition of symbolic music, and controlling a goal seeking agent in a dynamic environment, with further proposals for language driven localization and navigation in 3D environments.

 
Vous êtes cordialement invité.