Year of Graduation
Create Learning Algorithms to Play Computer Games Using OpenAI Platform
Fundamental Informatics and Information Technologies
The purpose of this work is to build and train several neural networks that can learn to successfully play several games. In the process of work, in order to obtain the most generalized solution only one architecture, method and parameters of training were used for each network. There was conducted a review of existing algorithms, their classification and comparative analysis. Therefore, two main approaches for problem solving have been studied: Evolutionary and Genetic algorithms for learning the neural network. On their basis, two algorithms were constructed, each of which solves two selected problems. During the work, the analysis of the solutions was carried out and made a comparative analysis of the two approaches. The ability to solve various computer games with one general architecture of the neural network demonstrates the consistency of self-learning algorithms for solving the tasks. This means that similar models can later made as a basis for algorithms that solve practical problems.