The Emergence of Diversity and Stability: from Biological Systems to Machine Learning

The observation of emergent properties of biological systems has been the inspiration of successful technologies opening new fields of computer science like artificial neural nets, swarm intelligence algorithms, evolutive algorithms, etc. In this work we focus on the emergence of negative feedback cycles: self-regulatory mechanisms able to react to alterations of some environmental parameters (temperature, gas concentrations, solar light, etc.) in order to compensate, preserving the environment in a state suitable for life. We make the hypothesis that speciation events play a central role for feedback formation and, and in order to select the negative cycles, the arising species need to be strongly connected to the environment, therefore the speciation needs to be sympatric (a speciation mode where new species arise without geographical isolation). As an intermediate result, we propose a simulative model of sympatric speciation and apply it to the field of evolutive algorithms. We propose some variations of the standard island model, a model used in evolutive algorithms to evolve multiple populations, to obtain dynamics similar to the sympatric speciation model, enhancing the diversity and the stability of the evolutive system. Then we propose a technique to define a metric and calculate approximated distances on very complex genetic spaces (a recurring problem for several evolutionary algorithms approaches). Finally, we describe the more complex model of negative feedback cycles emergence and discuss the problems that, in the current model formulation, make it not applicable to real world problems but only to ad hoc defined resource spaces; conclusively we propose possible solutions and some applications.