Fiche de MERCKLING Astrid

Titre : Unsupervised Pretraining of State Representations in a Rewardless Environment
Date : 22 septembre 2021
Heure: 14:00
Résumé : This thesis seeks to extend the capabilities of state representation learning (SRL) to help scale deep reinforcement learning (DRL) algorithms to continuous control tasks with high-dimensional sensory observations. SRL allows to improve the performance of DRL by providing it with better inputs than the input embeddings learned from scratch of the end-to-end approach. It involves solving state estimation in the manner of deep unsupervised pretraining of state embeddings without reward. These state embeddings must verify certain properties to allow for the correct application of bootstrapping and other decision-making mechanisms common to supervised learning, such as being low-dimensional and guaranteeing the local consistency and connectivity (or topology) of the environment, which we will seek to achieve through the models pretrained with the two SRL algorithms proposed in this thesis.
Lieu : ISIR , Campus Pierre et Marie Curie, 4 place Jussieu, Tour 65/66, 3eme étage, 75005 Paris