Towards Autonomous Reinforcement Learning: Automatic Setting of Hyper-parameters using Bayesian Optimization

BARSCE, JUAN CRUZ; PALOMBARINI, JORGE; ERNESTO MARTÍNEZ

Córdoba

Simposio; SLIOIA- Simposio Latinoamericano de Investigación de Operaciones e Inteligencia Artificial; 2017

SADIO

With the increase of machine learning usage byindustries and scientific communities in a variety of tasks such astext mining, image recognition and self-driving cars, automaticsetting of hyper-parameter in learning algorithms is a key factorfor obtaining good performances regardless of user expertisein the inner workings of the techniques and methodologies. Inparticular, for a reinforcement learning task, the efficiency of anagent learning a policy in an uncertain environment has a strongdependency on how hyper-parameters in the algorithm are set.In this work, an autonomous framework that employs Bayesianoptimization and Gaussian process regression to optimize thehyper-parameters of a reinforcement learning algorithm is proposed.A gridworld example is discussed in order to show howhyper-parameter configurations of a learning algorithm (SARSA)are iteratively improved based on two performance functions.