Survey of Deep Neural Networks Handling Plan Development using Simulations of Real-World Environments

Abstract

Neural networks are computational models that demonstrate the capability of advanced computing applications in plan development tasks. Five influential projects that independently demonstrate various applications of neural network models include AlphaGO/AlphaGO Zero, Playing Atari with Deep Reinforcement Learning, NeuroChess, OpenAI Five, and Playing Checkers without Human Expertise. Each of these projects includes different approaches to plan development tasks and are surveyed in this thesis using four criteria: the efficiency of the system, the form of their input based on their target environment, the structure of the neural network(s), and the processes through which they are trained. Each project approaches the plan development problem with strict regard to the environment in which they are targeting and thus vary in implementation. This survey is a collection of the details regarding each project and how the research teams approached their development