Hello everyone,
We have a semester project available at the Laboratory of Computational Neuroscience on Deep Reinforcement Learning for Autumn 2022 -- see the details below. The project is going to be supervised jointly by me and Dr. Johanni Brea.
Are you interested? Please ask me for more details in the next exercise session -- or send us your CV and grades (to alireza.modirshanechi@epfl.ch and johanni.brea@epfl.ch).
Have a nice rest of the day,
Alireza
Title: Tabular Exploration and Change-Point Detection in Deep Reinforcement Learning
Description: Discretization of continuous state spaces allows to use efficient methods developed for tabular settings. For example, it is possible to run efficient model-based reinforcement learning on learned discrete representations of the VizDoom environment (see Corneil et al. 2018). In this project we investigate how this can be further improved with modern approaches to exploration (Xu et al. 2021), change point detection (Liakoni et al. 2021) and state abstraction (van den Oord et al. 2017).
