Validation Methodology for AgentBased Simulations Workshop AgentBased Simulations
Validation Methodology for Agent-Based Simulations Workshop Agent-Based Simulations (ABS) Working Definition R. W. Eberth Sanderling Research, Inc. 01 May 2007
Purpose • Propose a definition of ABS for this workshop – not for the M&S community as a whole • For this workshop, widen the field of view to address other non-physical M&S classes that may require similar V&V approaches • Intent is to be Inclusive vice Exclusive
Ground Rules • For our purposes: § Agent-Based Simulation (ABS) and Agent. Based Modeling or Model (ABM) are synonymous § “Agent-Based, ” “Individual-Based, ” and “Entity-Based” are synonymous
Types of Models/Simulations • Physics-Based § Outcomes could be calculated, but math is too hard § Validity assessable empirically • Probability-Based § § May include physics Account for some/most random effects Range of outcomes could be calculated (see above) Validity assessable empirically • Agent-Based § § Emergent Behaviors Outcomes not predictable beyond small time increments Non-repeatable outcomes Validity assessable empirically?
Common to All Models/Simulations • May not account for all (or may misunderstand thus inaccurately account for): § Interactions § Cause and effect relationships • Account for no unknowns
Building Blocks of Definition • Multi-Agent System (MAS): any computational system whose design is fundamentally composed of a collection of interacting parts* • Individual-Based Models: simulations based on the global consequences of local interactions of members of a population (a subset of MAS)* • Complex Adaptive Systems (CAS): systems having the ability to self-organize and dynamically reorganize their components in ways better suited to survive and excel in their environments** § CAS Properties (J. Holland, 1995): • Aggregation: allows groups to form • Nonlinearity: invalidates simple extrapolation • Flows: allow the transfer and transformation of resources and information • Diversity: allows agents to behave differently from one another and often leads to the system property of robustness * Derived directly from the work of C. Reynolds ** As cited by D. Samuelson/C. Macal, 2006
Proposed Working Definition of ABS for This Workshop • Any computational system whose design is fundamentally composed of: § Autonomous decision-making entities (agents) § Agents interacting with each other and their environment over time § Agents independently sensing and responding to each other and their environment according to their own rule sets § Heterogeneous agent population § Agent decision-making algorithms that may be extremely simple or highly complex or that may evolve
• • • Other ABS Characteristics To Consider Large trajectory space “Edge of Chaos” outcomes Emergent behaviors Non-linearity Outcomes may be locally predictable (within a very small neighborhood), but are globally unpredictable • Whole emerges from the parts • Principal distinction between ABS and traditional analytic modeling: § Bottom-up versus top-down • “A synonym of ABM would be microscopic modeling, and an alternative would be macroscopic modeling. ” • Not “Can you explain it? ” but “Can you grow it? ” E. Bonabeau, 2002
For This Workshop Include: • Classes of non-physical, nonprobabilistic models, including: § § § Decision rule sets Human behavior representation (HBR) Knowledge-based systems (KBS) Cellular automata models Population dynamics Political, Military, Economic, Social Infrastructure and Information (PMESII) models § Social models
Related Issues (Only Touched on in This Workshop) • Optimization and Heuristics § Objective function formulation § Selection of one solution from many • Data § § Weighting factors Thresholds Breakpoints Others
- Slides: 10