University of Hull

CSRG - Agent-Based Modelling (ABM)

CSI

| CSI Technical Reports | Department of Computer Science | University of Hull |

Agent-Based Models (ABMs)

Misc ABMs

Agent-based modelling (ABM) has become a very powerful approach to studying complex systems that are not easily modelled using closed-form analytical mathematical techniques. Models such as financial markets, socio-economic systems, systems involving large numbers of individual agents such as humans or other simple components of a large system are all amenable to this approach. Generally, one formulates a very simple if-then-else model for an individual agent and this is implemented in some sort of software management framework so that a very large number of these simple individuals can interact. In many cases one finds complex emergent behaviour at a macroscopic scale that would be unanticipated from the simplistic individual rules defined at a microscopic level.

The art of successful ABM is often being able to rapidly prototype a model that is capable of sustaining a large enough set of individual agents. Conventional modelling packages can usually sustain a few tens or sometimes hundreds of agents. Bespoke optimised software is usually needed to manage millions or tens of millions of agents however. This is the sort of scale that one sees log-scaled emergent phenomena and is necessary to model phenomena such as financial markets, social or political systems, or even the Internet itself.

The composite tableaux above shows a selection of screen-dumps from simulated models using different lattices - Top row: Square 128x128; Game-Of-Life; Random Walk; Langton Ant, Middle row: Hexagonal 128x128; Epidemic; Ising; Reichenbach, Bottom row Triangular 128x80; Ising, Potts (Q=4); Forest Fire. We routinely simulate larger system sizes than these, but they become harder to render graphically when larger than those shown.

The individual microscopic components of a large scale complex system are often modelled as artificially intelligent spatial software agents. These animats occupy a simulated physical space and interact with one another in that space. There has been much success reported in the literature using relatively simplistic game theoretic models such as the Prisoners' Dilemma, the Snowdrift model, as well as wealth based models such as the SugarScape system in this manner. There are a number of financial, social or economic datasets that can be used for comparison with ABM simulated data.

The energy and renewables industry sector poses a number of interesting problems that can be formulated in terms of microscopic consumer agents, producer agents and distribution agents. This industry sector is based on a highly nonlinear and complex set of networks and is not easy to simulate using conventional and mathematical techniques. There is great scope to formulate realistically sized models for energy systems that would greatly aid regional and national scale planning and what-if scenario explorations.

The international and national financial markets are also amenable to an ABM approach and ABM is coming to be recognized as an important approach after the global failures of the theoretical pricing models that were based on stochastic differential equations. Other socio-economic systems such as political voting and opinion formation also offer scope for sophisticated models to aid regional and national sociological planning.

Managing the numerical experiments and analysis of statistical data from the use of a large scale ABM is one challenging area of computational science research. The problems of developing a robust and high performance ABM simulation framework are also important areas that can make use of both parallel and high performance computing technologies as well as sophisticated graphical rendering tools.

Agent-Based Modelling (ABM) is a broad area covering many different sorts of agents and indeed models. Generally ABM is not the same as agent-oriented artificial intelligence, although the two areas share some common ideas.

A particular interest is the class of spatial models that can be defined in terms of reflexive behaviour and from which complex emergent spatial patterns and behaviours arise on macroscopic scales, even when the individual microscopic agents are very simple in nature. We have studied these sort of models on a variety of geometries, dimensions, and with different interaction neighbourhoods or localities. There are many interesting quantitative metrics and statistical properties that can be studied through numerical experiments. The comptutational science investigation of ABMs comes from a careful posing of these numerical experiments in order to understand the emergence of both temporal and spatial properties. In particular, it is useful to investigate how simple individual agents can be formulated, while still retaining macrosopic complexity, and also in studying the sensitivity - or sometimes lack of it - to different agent starting configurations.

Although ABM systems can be interesting with relatively small numbers of agents, and these small systems of a few tens or a few hundreds of agents can be simulated with off-the-shelf software packages, our speciality is in studying very large systems of millions to billions of individual agents. On such system sizes there is scope for the emergence of macroscopic phenomena that can only be fully appreciated and quantified on logarithmic scales. To simulate such systems for interestingly long periods of time it is necesary to use sophisticated programming techniques to optimise the run time performance. This sometimes requires the use of High Performance Computing (HPC) or parallel computing on a massive scale. Visualising large scale agent systems is a lso a computational challenge, particularly when agent parameters need to be modified in interactive time.


Some Areas of Research Involving Agent-Based Modelling

The links below connect to Technical Notes or peer-reviewed and published articles on various areas of our agent-based modelling research activities. Links are to abstracts and bibliographic details, and in most cases also link onwards to a full-PDF version of each article.


Axelrod Model

CSTN-170: Dimensional and Neighbourhood Dependencies of Phase Transitions in the Axelrod Culture Dissemination Model

Conway's Game of Life

CSTN-151: Field Programmable Gate Arrays for Computational Acceleration of Lattice-Oriented Simulation Models

Daisyworld Model

CSTN-201: Dimensional Dependencies of Species Survival in the Daisyworld Wodel

Game of Death Model

CSTN-130: Cycles, Transients, and Complexity in the Game of Death Spatial Automaton

Diffusion-Limited Aggregation

CSTN-107: Morphology and Epitaxial Growth in a Directed Diffusion Model

Diffusion-Limited Cluster-Cluster Aggregation

CSTN-175: Characterising Components and Flocculation Structure of Sediment in a Diffusion-Limited Cluster-Cluster Aggregation Model
CSTN-012: Simulating and Visualising Sedimentary Cluster-Cluster Aggregation

Eden Model

CSI-0062: Simulations of Structural Asymmetries and Dimensional Dependence of Scaling in the Screened Eden Model
CSTN-086: Dynamical Runaway Growth and Simulation of Cancer amongst Spatial Animat Agents

Epidemic Model

CSI-0003: Role of Connectivity and Clusters in Spatial Cyclic SIRS Epidemic Dynamics

Forest Fire Model

CSTN-163: Transients in a Forest-Fire Simulation Model with varying Combustion Neighbourhoods and Watercourse Firebreaks

Generalised Game of Life

CSTN-137: Static and Dynamical Equilibrium Properties to Categorise Generalised Game-of-Life Related Cellular Automata

Hawk and Dove Model

CSTN-150: Well-Mixed Systems and the Approach to Equilibria in Spatial Hawk and Dove Game Simulations

Invasion Percolation Model

CSTN-134: Gravitational and Barrier Effects in d-Dimensional Invasion Percolation Reservoir Models

Ising Model

CSTN-004: Agent Formulation of the Ising Model
CSTN-036: Ising Model Scaling Behaviour on z-Preserving Small-World Networks
CSTN-104: Cluster and Fast-Update Simulations of Regular and Rewired Lattice Ising Models Using CUDA and Graphical Processing Units
CSTN-108: Visualising Spins and Clusters in Regular and Small-World Ising Models with GPUs
CSTN-148: Hybrid Update Algorithms for Regular Lattice and Small-World Ising Models on Graphical Processing Units

Kawasaki Model

CSTN-147: Visual Simulation of a Multi-Species Coloured Lattice Gas Model
CSTN-132: Visualising Multi-Phase Lattice Gas Fluid Layering Simulations
CSTN-109: Data-Parallelism and GPUs for Lattice Gas Fluid Simulations

Potts Model

CSTN-135: Bit-Packed Damaged Lattice Potts Model Simulations with CUDA and GPUs

Prisoners' Dilemma Model

CSTN-081: Roles of Space and Geometry in the Spatial Prisoners' Dilemma
CSTN-037: Non-Monotonic Phase Transition Edges in the Spatial Prisoners' Dilemma

Rock-Paper-Scissors Model

CSTN-066: Complex Domain Layering in Even-Odd Cyclic State Rock-Paper-Scissors Game Simulations

Rock-Paper-Scissors-Lizard-Spock Model

CSTN-129: Cycles, Diversity and Competition in Rock-Paper-Scissors-Lizard-Spock Spatial Game Simulations

Schelling Model

CSI-0001: Multiple Species Phase Transitions in Agent-Based Simulations of the Schelling Segregation Model
CSTN-198: Multiple Species Effects and Transitions in Schelling Segregation Agent-Based Model Simulations
CSTN-182: Neighbourhood, Boundaries and Population Dependencies in Schelling Segregation Model Simulations

SIRS Model

CSI-0003: Role of Connectivity and Clusters in Spatial Cyclic SIRS Epidemic Dynamics

Snowdrift Model

CSTN-133: Cost Benefits and Cooperation in Spatial Snowdrift Game Agent Systems Approaching Well-Mixed Equilibria

Sugarscape Model

CSI-0006: Neighbourhood, Geometry and Initial Conditional effects on Fairness and Agent Longevity in Large-Scale SugarScape Models

Sznajd Model

CSTN-144: Halo Gathering Scalability for Large Scale Multi-dimensional Sznajd Opinion Models Using Data Parallelism with GPUs
CSTN-032: Multi-Party and Spatial Influence Effects in Opinion Formation Models

Volunteer Model

CSI-0004: Neighbourhood Size Dependence of Cooperator Cluster Emergence in the Spatial Volunteers' Dilemma

Tools and Simulation Infrastructure for ABMS

CSI-0027: 3D Print Technology for Cellular Agent-Based Growth Models

CSI-0010: Automatic High Performance Structural Optimisation for Agent-based Models

CSTN-209: Use of Closures to Engineer Software for a Family of Numerical Simulation Models
CSTN-113: Multiphase Updating - A Practical Approach to Simulating Animat Agents

ABM Applications

CSTN-142: Software Engineering a Family of Complex Systems Simulation Model Apps on Android Tablets
CSTN-116: An Agent-Based Model of the Battle of Isandlwana

Animat Predator-Prey Formulated ABM Applications

CSTN-188: Emergent System Effects from Microscopic Evasion Choices in a Predator-Prey Simulation

CSTN-184: Modelling Predator Camouflage Behaviour and Tradeoffs in an Agent-Based Animat Model

CSTN-178: Introducing a Gestation Period of Time-Delayed Benefit into an Animat-based Artificial Life Model

CSTN-121: An Investigation into the Effects of Sentinels on Animat Collectives

CSTN-097: Simulating Intelligent Emergent Behaviour amongst Termites to Repair Breaches in Nest Walls

CSTN-094: Emergent Societal Effects of Crimino-Social Forces in an Animat Agent Model

CSTN-090: Elucidating Soldier and Worker Caste Divisions in an Animat Artificial Life Model

CSTN-086: Dynamical Runaway Growth and Simulation of Cancer amongst Spatial Animat Agents

CSTN-085: Cross-Caste Communication in a Multi-Agent Predator-Prey Model

CSTN-084: Spatial Animat Agent Evolution and Changing Ecological Niches

CSTN-078: Spatial Pattern Growth and Emergent Animat Segregation

CSTN-076: Intelligent and Adaptive Animat Resource Trading

CSTN-068: Quantifiable Metrics for Complex Emergence in Spatial Agent-Based Models

CSTN-067: Complex Emergent Behaviour from Evolutionary Spatial Animat Agents

CSTN-060: Animat Swarms and Spatial Emergence Phenomena

CSTN-059: Resource Scarcity Effects on Spatial Species Distribution in Animat Agent Models

CSTN-056: Emergent Spatial Agent Segregation

CSTN-055: Pack-Hunting Multi-Agent Animats

CSTN-050: Energy Flow and Conservation in an Artificial Life Agent Model

CSTN-047: Global Constraints and Diffusion in a Localised Animat Agent Model

CSTN-045: Hierarchical Relationships and Spatial Emergence Amongst Multi-Species Animats

CSTN-044: Altruism Amongst Spatial Predator-Prey Animats

CSTN-041: Boundary Conditions and Locality in an Agent-Based Predator-Prey Model

CSTN-040: A Minimal Spatial Cellular Automata for Hierarchical Predator-Prey Simulation of Food Chains

CSTN-035: Spatial Emergence of Genotypical Tribes in an Animat Simulation Model

CSTN-033: Tools and Techniques for Optimisation of Microscopic Artificial Life Simulation Models

CSTN-027: Grid-Boxing for Spatial Simulation Performance Optimisation

CSTN-020: Roles of Rule-Priority Evolution in Animat Models

CSTN-017: Manual and Semi-Automated Classification in a Microscopic Artificial Life Model

CSTN-015: A Zoology of Emergent Life Patterns in a Predator-Prey Simulation Model

CSTN-010: Defensive Spiral Emergence in a Predator-Prey Model

CSTN-009: Parallel Synchronisation issues in Simulating Artificial Life

CSTN-007: A Framework and Simulation Engine for Studying Artificial Life

| CSI Technical Reports | Ken Hawick | Department of Computer Science | University of Hull |