An ecosystemic view for developing biologically plausible optimization systems

Abstract

The search for plausible biologically inspired ideas, models and computational paradigms always drew the interest of computer scientists, particularly those from the Natural Computing area. Also, the concept of optimization can be abstracted from several natural processes, for instance, in the evolution of species, in the behavior of social groups, in the dynamics of the immune system, in the food search strategies and in the ecological relationships of different animal populations. To the best of our knowledge, ecosystems and their concepts have not been explored computationally in the context of function optimization and, therefore, they are addressed in this thesis. This work highlights the main properties of ecosystems that can be important for building computational tools to solve complex problems. Also, it is introduced the computational modelling for such biologically plausible functionalities (e.g., habitats, ecological relationships, ecological succession, and another). The main discussion presented in this work relates to the cooperative use of populations of candidate solutions, coevolving in an ecological context. With this ecology-based analogy, each population can behave according to a specific search strategy, employed in the evolution of candidate solutions. In addition to the possibility of using different optimization strategies cooperatively, this analogy opens the possibility of inserting ecological concepts in the optimization process, thus allowing the development of new bio-plausible hybrid systems. The potentiality of some ecological concepts is also presented in a canonical Ecology-inspired Algorithm for Optimization, named ECO. Some population-based algorithms are used to compose the ecology-based approach. The problems solved in this thesis are several continuous benchmark functions with a high number of dimensions ($D = 200$), and the protein structure prediction problem for the 2D AB model. Also, the use of population dynamics to self-regulate the size of populations during ecological successions; the use of heterogeneous models embedding different search strategies into the system; and the use of hierarchical clustering to dynamically adjust the habitats formation and probabilistically define the habitats communication are some case studies investigated. Results were promising concerning the application of the proposed computational ecosystem for optimization. Finally, concluding remarks and several ideas for future research are presented.