Simulation of the Unified Learning Model Using Repast

Introduction

The Unified Learning Model is a model of how people learn and a resulting model of teaching and instruction. It is a unifying synthesis of theories on how working memory, knowledge, and motivation work together when we learn. It is founded on three basic principles of learning: (1) learning is a product of working memory allocation, (2) working memory’s capacity for allocation is affected by prior knowledge, and (3) working memory allocation is directed by motivation (Shell et al. 2010).

How is this related to Intelligent Agents and Multiagent Systems? This Unified Learning Model can provide an algorithmic framework for metareasoning in agents, if one maps working memory to computational resources for reasoning processes, knowledge to data and information that an agent has, and motivation as the goals and intentions of an agent. This framework is more comprehensive than the popular Belief-Desire-Intention (BDI) model (Rao and Georgeff 1995) because it considers the working memory and the learning process – the acquisition of new data and information and how that impacts the change in one’s reasoning. And thus, I see potential extension to the BDI model with this unified model.

So the simulation could be based on the theoretical extension of the BDI model, or implementation of the unified learning model in Repast (a simulation software), or both.

The impact of this work could be significant. First, it is novel as it considers learning and teaching in agents in the most basic way: how working memory works with knowledge and motivations, where knowledge is more general, more explorative while motivation is more targeted, more exploitation. Second, it could extend the popular BDI model significantly theoretically, defining a framework that allows MAS researchers to consider and define how agent learning and teaching should be done. Third, the simulation software could be shared and become a vital tool for MAS researchers investigating agent interactions and meta-reasoning on an individual basis as well as on a “group” basis.

Repast Simulation Software

Check out http://repast.sourceforge.net/index.html for a detailed description of the testbed. There is also FAQ at the website – please review them carefully. Notice that you will need to use Repast 3 version for your project. A detailed demonstration of Repast environment will be given in the classroom on September 24, 2009. In addition, a detailed user manual for running the Repast simulation toolkit in the Netbeans programming environment is available at: http://cse.unl.edu/~knobel/ma-simulation/

Briefly, as reported on the website: The Recursive Porous Agent Simulation Toolkit (Repast) is one of several agent modeling toolkits that are available. Repast borrows many concepts from the Swarm agent-based modeling toolkit. At its heart, Repast toolkit version 3 can be thought of as a specification for agent-based modeling services or functions.

Repast 3 has a variety of features including the following:

· Repast includes a variety of agent templates and examples. However, the toolkit gives users complete flexibility as to how they specify the properties and behaviors of agents.

· Repast is fully object-oriented.

· Repast includes a fully concurrent discrete event scheduler. This scheduler supports both sequential and parallel discrete event operations.

· Repast offers built-in simulation results logging and graphing tools.

· Repast has automated Monte Carlo simulation framework.

· Repast provides a range of two-dimensional agent environments and visualizations.

· Repast allows users to dynamically access and modify agent properties, agent behavioral equations, and model properties at run time.

· Repast includes libraries for genetic algorithms, neural networks, random number generation, and specialized mathematics.

· Repast includes built-in systems dynamics modeling.

· Repast has social network modeling support tools.

· Furthermore, the Intelligent Agents and Multiagent Systems (IAMAS) Group at UNL has built a suite of packages (search algorithms, case-based reasoning, probabilistic modeling and reasoning, etc.) for Repast and will be made available for the class to use.

References

Shell, D. F., D. W. Brooks, G. Trainin, K. M. Wilson, D. F. Kauffman, and L. M. Herr (2010). The Unified Learning Model, Springer.

Rao, A. S. and M. P. Georgeff (1995). BDI-agents: From Theory to Practice, In Proceedings of the First International Conference on Multiagent Systems (ICMAS'95), San Francisco, CA.