Social learning supported deep rainforcement learning

Social learning is the process by which individuals acquire information from other individuals in their environment and use this information to shape their behavior. This thesis presents a novel Deep Reinforcement Learning algorithm that supports social learning inspired by the copying mechanism in nature using intrinsic feedback. The aim is to increase the learning speed and performance of individuals within a group. Deep Q learning, a well-known reinforcement learning algorithm, is used. Compared to other research using imitation with reinforcement learning, our approach is novel in that the learning agents are controlled by a dual-layer control system, different types of memory are examined, and copying and application behaviors are dynamically controlled by the intelligent agent by defining the actions that the deep neural network can perform according to its nature. Our approach is implemented in a simulation environment. The results of the simulation experiments show that the learning speed increases and the copy command as an action has a positive effect.