Survival by continuous learning in a dynamic multiple task environment

The ability to adapt or not when challenged with a dynamic, changing environment is what differentiates between survival and extinction of species. In this paper we present a machine learning method that allows agents in an environment with changing tasks to adapt and modify their behavior thus ensuring their survival. These agents do not get explicit information about the change in tasks. The learning mechanism ensures the presence of enough diversity in the agents so that they can restart learning if the previous learning stops to be effective and enough continuity in the system so that the agents can keep on learning if their task has not changed.