Experimental Trials Based on a Neocortex-based Adaptive System Pattern

Abstract

This paper proposes a general design pattern for building adaptive systems. The Neocortex Adaptive System Pattern (NASP) architecture is an adaptive decision-making architecture. It is derived from the physical architecture observed within the neocortex of a primate brain. This architectural pattern is used as a basis to provide necessary functions to adaptive systems, allowing different adaptive system components with different methodologies and techniques to coexist and cooperate within a single system. Properties of the NASP are illustrated using an agent-based simulation experiment framework composed of simulated tank vs. tank game. This study supplies experimental results that compare adaptive decisions based on accuracy and timeliness. It shows that a more accurate decision may in fact be the less optimal one due to time constraints. The experimentation results suggest that multi-system adaptation can increase system performance, and learned information can identify time frames when an adaptation can increase system performance. The practice of designing and building agent based systems shares many principles and approaches with the NASP. An agent-based architecture has a common environment that is utilized to share the state of the system with member agents. It contains autonomous entities that communicate with each other in order to perform their designed functions. A unique contribution of the NASP approach over other research is to add the ability for different agents to create alternative courses of action and controls such as rule-based, neural, or Bayesian that are used to choose from those alternatives based on their latest information. While counter intuitive, the findings suggest that increased performance in this combatant domain suggest that earlier adaptations, using less information, improve the performance of the adaptive system. The paper provides a literature review of relevant neuroscience literature that describes the parallels between the architecture of the neocortex and NASP. The paper discusses the simulation experiments and associated results that illustrate how tradeoffs between information completeness and timeliness affect system performance within a NASP-based system.