| dc.description.abstract | This thesis presents a novel approach for solving reinforcement learning tasks using elementary cellular automata (ECA) based reservoir computing (RC). Combining theories from these three fields to create a baseline for further investigation. The main objective is to investigate the unique ECA rules and their properties, and test them in different reinforcement learning (RL) environments. ECA based RC have previously been tested on the x-bit memory benchmark, where it has shown capabilities of long short-term memory. For many RL tasks, memory has been shown to be a vital part for good performance. This, together with the reservoirs computationally efficient methods of learning, makes this a quick method for training high performing models. In the experiments performed, it is shown that variations in the reservoir size, number of iterations, number of cells updated and how they are updated can have a huge impact on performance when certain rules are used. | en_US |
