N. Boyko , B. Dokhniak, V. Korkishko
Abstract. This article is devoted to the algorithm of training with reinforcement (reinforcement learning). This article will cover various modifications of the Q-Learning algorithm, along with its techniques, which can accelerate learning using neural networks. We also talk about different ways of approximating the tables of this algorithm, consider its implementation in the code and analyze its behavior in different environments. We set the optimal parameters for its implementation, and we will evaluate its performance in two parameters: the number of necessary neural network weight corrections and quality of training.
Keywords: Training with reinforcement, Q-Learning, Neural networks, Markov environment.