Using a 'Naturalistic' Mechanism to Capture Cognition and Dynamic Behaviors in Task Network Models: An Overview

摘要

Over the past several years, we have developed and tested a computational implementation of recognition-primed decision making (RPD) within Micro Saint Sharp task network models. The goal of this work was to augment task network models to improve their cognitive fidelity and flexibility. Our RPD mechanism uses multiple trace memory with a simple reinforcement-based learning to pick up on the statistics of a model environment. Here we present a collection of models and validation results that include classic categorization and probabilistic category learning tasks, dynamic adversarial and control tasks, and a larger-scaled militarily-relevant task. The range of tasks we have modeled demonstrates the flexibility of our approach and what can be accomplished with simple mechanisms.