30 Jul 2021
14:00 Doctoral defense Fully distance
Theme
Evocycle: artificial life in an ecosystem of bio-inspired algorithms
Student
Felipe dos Santos Pinto de Andrade
Advisor / Teacher
Ricardo da Silva Torres
Brief summary
Artificial life is the field of study that aims to reproduce behaviors analogous to life in different types of media. In the soft approach, this task is performed using software simulations. This thesis presents an artificial life framework, called Evocycle, which uses different bioinspired meta-heuristics as system agents. Such meta-heuristics are part of the field of Evolutionary Computing in Computer Science, in which the biological theory of evolution is taken as inspiration to design stochastic optimization algorithms. In classical evolutionary computing, elitist approaches are often used, such as using fitness functions to select individuals for new generations. However, natural selection as observed in nature does not have a fitness function for evaluation. In this work, we propose a framework that removes the emphasis of evolution on fitness values, and tries a different approach, based on ecological relationships of feeding, reproduction and predation. Our artificial life system forms an ecosystem in which different species are modeled by the hybridization of collective and evolutionary intelligence algorithms, and the interaction of these species through ecological relationships builds their behavior over time. To validate the proposed framework, we simulate it with implementations using Particle Swarm Optimization (PSO) algorithms, in which the velocity function of each particle is defined by a genetic programming tree (GP), and the Artificial Bee optimization algorithm. Colony (ABC). Rather than a complete cycle of fitness function selection, the GP trees, which encode each PSO particle, evolve based on survival within the simulation, taking into account factors such as internal energy, age, and proximity to other individuals. This ecosystem was studied in continuous optimization benchmarks, and by analyzing the genetic frequency variation of populations of interest we observed the development of interesting ecological dynamics. We also assess the resilience of the predator population in the system using studies from disaster theory in biology. The main contributions of this work are the proposal of the artificial life system itself, the study of the emergence of the search behavior of the system agents, and the proposal to use the analysis of recovery rates after disturbances in the population to assess its resilience.
Examination Board
Headlines:
| Ricardo da Silva Torres | IC / UNICAMP |
| Ulisses Martins Dias | FT / UNICAMP |
| Guilherme Palermo Coelho | FT / UNICAMP |
| Rafael Stubs Parpinelli | CCT/UDESC |
| Douglas Rodrigues | FC / UNESP |
Substitutes:
| Fábio Luiz Usberti | IC / UNICAMP |
| Luiz Camolesi Júnior | FT / UNICAMP |
| João Paulo Papa | FC / UNESP |