Reinforcement Learning by Comparing Immediate Reward - Computer Science > LearningReportar como inadecuado

Reinforcement Learning by Comparing Immediate Reward - Computer Science > Learning - Descarga este documento en PDF. Documentación en PDF para descargar gratis. Disponible también para leer online.

Abstract: This paper introduces an approach to Reinforcement Learning Algorithm bycomparing their immediate rewards using a variation of Q-Learning algorithm.Unlike the conventional Q-Learning, the proposed algorithm compares currentreward with immediate reward of past move and work accordingly. Relative rewardbased Q-learning is an approach towards interactive learning. Q-Learning is amodel free reinforcement learning method that used to learn the agents. It isobserved that under normal circumstances algorithm take more episodes to reachoptimal Q-value due to its normal reward or sometime negative reward. In thisnew form of algorithm agents select only those actions which have a higherimmediate reward signal in comparison to previous one. The contribution of thisarticle is the presentation of new Q-Learning Algorithm in order to maximizethe performance of algorithm and reduce the number of episode required to reachoptimal Q-value. Effectiveness of proposed algorithm is simulated in a 20 x20Grid world deterministic environment and the result for the two forms ofQ-Learning Algorithms is given.

Autor: Punit Pandey, Deepshikha Pandey, Shishir Kumar


Documentos relacionados