Effect of Hyper-Parameter Tuning on the Performance of Augmented Random Search

Recently, researchers working in the field of Artificial Intelligence have been keen on finding new algorithms that provide optimal results. One such reinforcement learning algorithm is the Augmented Random Search. This algorithm is imperative as it provides better results over other reinforcement learning algorithms. Finding the optimal values of hyper-parameters helps the algorithm to work up to its full potential. In this work, we have made an effort to find out the best hyper-parameter settings for the learning rate and direction, for this specific task. Our results show how a change in these values can impact the final maximum rewards obtained during the training process. We report novel findings of the impact of hyper-parameters on the performance of ARS algorithm. Finally, an effort has been made to find out by what percentage the ARS performs better than other Reinforcement Learning algorithm, TRPO.