Group Interests of Agents Functioning in Changing Environments

We consider agent populations with breeding split into groups with a naive common goal to survive in changing environments. In order to grasp principal tendencies in cooperation, the agents are modeled as very simple systems, the single layer perceptron (SLP) based classifiers. They ought to learn how to train themselves rapidly, adapt to unexpected pattern recognition task changes, to comply the fitness function and survive. Failure to comply survivability condition will result in the agent being replaced by a newborn that inherits some upbringing information from one of successful agents in the group. We found that inherited fraction of incorrect training directives (a noise) which controls the agent's ability to adapt to changes is following environmental alterations. Restricted cooperation between agent groups is beneficial to overcome outsized changes.