The Language and Voice Lab and CADIA present a PhD thesis lunchtime talk by Hlynur Davíð Hlynsson: Visual processing in context of reinforcement learning.
Monday April 4th at 12:20
In this work, Hlynur presents three different representation learning algorithms in the context of reinforcement learning (RL): (i) GrICA is inspired by the Independent Component Analysis (ICA) and trains a deep neural network to deliver statistically independent components of input. (ii) Latent Representation Prediction (LARP) learns state representations by predicting the representation of the next state of the environment using a current state and a current action. (iii) RewPred learns state representation by training a deep neural network to learn a smoothed version of the reward function. Each method has its strengths and weaknesses, and we conclude from our experiments that the inclusion of unsupervised representation learning in RL problem-solving pipelines can accelerate learning.