.png){kind=icon}
.png){kind=icon}
Return
Evolutionary self-optimization of large CA-based multi-agent systems
.png){kind=icon}
PRE
AI
DOI：10.1016/j.jocs.2023.101994.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
The paper is an extended version of Seredynski et al. (2022) [1] and presents a theoretical framework called a Competitive Co-evolutionary Cellular Automata-based System enabling self-optimization of large distributed systems. It contains three components: (a) a multi-agent, social-like interpreted system, where modeling of agents, discrete space and time is provided by a 2-dimensional Cellular Automata, (b) an interaction between agents described in terms of the Spatial Prisoner's Dilemma game, and (c) a local evolutionary mechanism of competition between agents. It is assumed that a multi-agent system is composed of several types of agents, sometimes also called species. The goal of each agent is to maximize its own payoff. Agent-species compete for space applying locally a mechanism of evolutionary selection, where the payoff is considered as fitness. As a result of the competition, more profitable species replace less profitable. While agents act locally to maximize their incomes, we study conditions of emerging of global collective behavior measured by the average total payoff of players of which they are not aware of. We show that collective behavior can emerge in a fully distributed way based on achieving a Nash equilibrium if some conditions are fulfilled. We also show the significance of the introduced income sharing mechanismin the self-optimization process providing a high level of global collective behavior.
Keywords:
Cellular automata
Collective behavior
Competition
Income sharing
Multi-agent systems
Self-optimization
Spatial Prisoner?s Dilemma game
Journal
IF:
18.3
Papers:
3.1K
Citations:
4.0K