SELF-MOTIVATED AND TASK-ORIENTED, MULTI-DIMENSIONAL LEARNING IN A DYNAMIC AND UNCERTAIN ENVIRONMENT WITHOUT HUMAN INTERVENTION

摘要

The abilities to accept new information from the environment and use it to update our existing knowledge thus adapting to the changes of our environment have played a crucial role in the success of human beings as a species. Incorporating these abilities in machines has been an age long desire of artificial intelligence. In this paper, we present a learning technique based on evolutionary approaches that enables artificial agents to detect changes in their environment and adapt accordingly. Our focus is on enabling the agents to learn new tasks without any human intervention, relying only on stimulus from their environment. We argue that learning in such a dynamic environment should be a continuous process and past experiences must be retained for future scenar ios. The learning method itself provides a mechanism where the decrease in performance, forced by the change in goals, triggers new learning. We conduct experimentation to show how this approach works and results from these experiments are very encouraging.