MIS 643 AgentBased Modeling and Simulation 20172018 Fall
MIS 643 Agent-Based Modeling and Simulation 2017/2018 Fall Chapter 1 Intorduction
Outline 1 Introduction 2 Models 3 From Simulation to Social Simulation 4 Agemts 5 Agent-based Modeling and Simulation 6 Applications 7 Resources 8 Conclusions
1 Introduction • Agent-based Modeling and Simulation (ABMS) – Paradigm, methodology – Modeling approach – aim – better undertand natural, socio-technical phenomena • agents – – – autonomous having properties and actions (behavior) individual heterogeneity interactive with other agent and their environments boundadly rational - adaptation and learning behavior • emergence of structure – macro or social levels • ABM - Computational modeling – Constucting models – a phenomena is modeled in terms of its agents, environment and their interactions • create, analyze, experiment with
Aim of the Course • • ABM – transformative representational technology better uderstand familiar topics make sense of and analyze – hiterto unexplained topics Developing ABM literacy – powerful, professional and life skill • Restructuration – from one structuration of a domain to another – change in representational infrastructure • E. g. : from Roman to Hindu Arabic numerals in Europe – dificult to reprent large numbers and performe aritmetic operations • E. g. : transformation of kinematics from vorbel to algebra
2 Models • Models – Building simplified representations of the phenomena • social, natural, business or socio-technical • Types of models: – Verbal - Natural languages – Analog – Mathematical – equation-based • Analytical – closed form solutions • Emprical: regression equations, neural networks – Single or SEM – interraction among variables – A relation between dependent and independent variables is estimated from data • Differential / difference equations (System dynamics) • Computational method – Computer programs – Inputs (like independent variables) – Outputs (like dependent variables)
Example of a Model • Consumer behavior model: – How friends influence consumer choices of indivduals • Buy according to their preferences • what one buys influences her friends decisions – interraction • verbal • Mathematical – Theoretical model – Emprical : statistical equations • estimated from real data based on questioners • Simulation models of customer behavior – ABMS – interractions, learning, influence from networks
Mathematical Models • Analytical models – closed form solutions • Restrictive assumptions – Rationality of agent – rational choice theory – Representative agents – Equilibrium • Contradicts with observations – Labaratory experiments about humman subjects – Observations at macro level – stylized facts • as precision get higher explanatory power lower • Relaxation of assumptions – geting a closed form solution is impossible
Example: Consumer behavcior • Consumer behavior models in economics • treat a typical consumer as a untility maximizing agent • the consumer observes prices of goods/services • derives utiity from them • perfectly rational • Mathematical tools – at minimum - calculus • Interraction of consumers in a market • two or three types of consumers • equilibrium is assumed
Emprical Models • Estimation of parameters of a single or set of equations from real world data • Methods – statistics, machine learning or data mining – Regression – single equation or SEM – Nueural networks, SVM, – Decisio trees • E. g. : estimate behavior of cunsumer from opinion survays • E. g. : behavior of an economy – Simultaneous equations
3 From Simulation to Social Simulation • Simulati. on: Model of a system with suitable inputs and observing the corresponding outputs • Uses of simulation Axelrod(1997) – – – – 1 -Prediction: 2 -Performance: 3 -Training: 4 -Entertainment: 5 -Education: 6 -Proof 7 -Understanding - Discovery:
Third Disipline • Inductive – Discovery of patterns in emprical data – E. g. : analysis of opinion data, econometirc models • Deductive – Axioms – assumptions – Proving consequences – theorems – E. g. : proving Nash equilibrua in games • Simulation – set of assumptions but not prove theorems – generates data – analyzed inductively • anaysis of simulation outputs • comparing with real data
Computational • use computers or ICT as an instrument • other examples instuments restructuring science – optical telecope - astronomy – microsope – bioloy – find other insruments restructuring sciences • Compare – Output of the model and data from real world – if output model is similar to real world • Validity of the model
Experiments • Experiment: – Applying some treatment to an isolated system and observing what happens • Common in natural sciences – Physics, chemistry • Not common in social sciences – isolation – Mostly in psychology, new in experimental economics • Computer simulations – chaning parameters - range – other factors randomly • if the model is a good representation of the reality – Senario or what if analysis
Simulation in Social Science • In engineering or natural science – Prediction – E. g. : predict – position of planets in the sollar system – motion of molecules – weather temperature (next day, hour) • In social science – – Uderstanding social phenomena, processes or mechanizms Proof of my claim or hypotheis Discover some new previously unknown patterns Policy/senario analysis
How to communicate • Induction – Publich model (equeation , coefficients, significance) • Deduction – Theorems, equeations • Simulation – Publish the pseudocode or algorithm – Outputs: graphical , plots, tables – Fit equations to the data generated by simulation
4 Agents • Distributed Artifical Inteligence (DAI) or multi-agent systems (MAS) • Agents - software – Searching internet: softbots, visards for assistance • Agents represents in ABMS – Individuals – consumers, producers, families – Organizations – governemts, merket makers – biological entities – animals, forest, crops • What they do – – – Get information from their environment or from other agents Process information, may have limited memory - forget Communicate with one onother via messaging Learn from others, their own experiences Try to atchive goals
Chacteristics of an Agent in MAS • • Multi-agent Systmms – branch of AI Four characterisitcs Woodridge & jannings, 1995) Autonomy Social ability – interract with other agents or humans (users) • Reactivity – React to stimula comming from its environment • Proactivity – Goal or goals
5 Agent based Modeling and Simulation • After – Modeling – Simulation – Agents • ABMS: – A simulation paradigm used in social and natural sciencees to analyze or better understand these sysems consisting of autonomous, interaction, goal-oriented and boundadly rational actors so called agents situated in an environment.
Complex Adaptive Systems • Complex systems - informally – difficult to understand – world we live getting more and more complex • many complex interractions compared to past • as science and technology progress • Simple to complex systems • Defined: • Systems with interracting many elements yet aggregate behavior can not be predictable from individual elements – from interractions of individual elements – an emergent phenomena arises • E. g. : population dynamics – Simple population dysnamics - all members are the same homogenous – complex food web - how each member interact with others
Complex Adaptive Systems • Properties Holland 2014 – Self-organized – order at the macro level – Chaotic behavior: small change in initial condition hase huge effects on system out – Fat-tailed: extream values more then normal distibution – Adaptive interactions.
Emergence • • • large scale effects of laocal interractions lower level to higher assumptions may be simple consequences may not be obvious –suprising micro level macro level phenomena micro – Second order emergence
Understaning Complex Systmes and Emergence • Two funamental and distict challenges • Integrative understanding – Try to figure out the aggregate pattern when knowing the indivdual behavior • Differential understanding – – The aggregate pattern is known Find indivdual behavior for that pattern Flocking behavior of birds V flocking of gooses
ABM - CAS • new computer technologies • simulate behaviors of interactiing agents • better uderstand arising complex patterns of natural and social systems • Alternative approcah - use simplified representations of complexity – sophisticated mathematical models • ABM computational methodology enableing modeling complex systems
Building Agent based Models • Problem • Agents – Cognitive and sensory charcteristics of agents – The actions they can carry out • Environment • Modeling – Conceptual model – – Implementation - programming Initial configration of the system Run the model Experimental setup • Observe the outcome – Often an emergent phenomena is looked for – Metamodel responce surface
A Generic ABM Simulation replication • Initialization – – clear all memory set time 0 creatre amd initilize agents set environmet parmeters • Repeat – increment time by one – at each time step • pass over all or some agents • perform some action • collect data • present data • until a stoping criteria • calcuate more statistics or outputs • present outputs
Model Development • Implementation of the model – simulate the model • • Varification Validation Analysis of the model Model development is an iterative process starting with problem formulation firet simple models get complicated
Validity • external – opperational validity • accuricy or adequecy of the model in matching the real world data – experimental, archivial, survay • Point prediction – natural systems • pattern predictions rubost processes – sequence of events similar not identical • Artificial societies – Artificial merkets – Abstract not real systems
Modeling Agents in ABM • Agents – – Reciving input from the environment Storing historical inputs and actions Actions and Distributing output • Symbolic AI – Production systems • Non symbolic – learning: adapting to changes – neural networks – evolutionary algorithms such as genetic algorithms • Object-oriented Programming
Object Oriented Programming • • • Classes – prototypes for each agent type Objects – agents - instances from each class Characteristics of agnet - Instance variables Behavior - Methods Interraction between - Mesage sending Inheritance/Polymorphism – from general agents to specific onces • Heterogenous in – characteristics – behaive differently
Software • High level languages – object oriented – Java, C++, C# • Special packages – – Swarm Repast Net. Logo MASON
The Agent’s Environment • Agents are in social environment – Network of interractions with other agents • Similar in characteristics – Physical – locations • Neighbour • Cellular autometa – Interract only with their claose neighbours
Features of ABM • Ontological correspondence – Computational agents in the model – real world actor – Desing the model, interpret results • Heterogenous agents – Theories in economics – actors are identical – Preferences, rules of behavior are different • Representation of the environment • Agent ınteractions • Bounded rationality – Optimizing utility v. s. limited cognitive abilitiesi • Learning – İndividual, population social levels
Adventages • Micro level macro level phenomena micro – Second order emergence • Programming languages – more expresive then mathematical models – modular: object oriented approach • No sofisticated mathematiical skills • Thought experiments – policy evaluation, senatio analysis • Enables to test different theories or hypothesis about a phenomena – E. g. : different consumer behavior theories
Limitations • Expresing the results – particular example • Rsults depends on – parameters – initaal conditions • Model communication – reproducibility of results – use standard packages – limitaitons • Interdiciplinary nature • Education in social science – no programming courses • May need computing power
Simulation Methods in Social Science • Gilbert(2005) classification – – – System dynamics Discrete event simulation – quing models Multilevel Microsimulation Cellular autometa Agent-based Simulation
Other Related Modeling Approaches • System dynamics (SD) • SD ABM : aggregate individual top- down buttom-up differential equations interacting agents • E. g. : Population dynamics • SD: a single variable for population – an equation describing its rate of channge – hard to include heterogenouty • ABM: modeling population with heterogenous agents – fertatlty, migration or death rate depends on – age, gender, income, etnicity, location
SD v. s. ABM (cont. ) • E. g. : population dynamics • E. g. : predator-pray • E. g. : technology diffusion
Microsimulation v. s. ABM • Microsimulation – – – Large database – individuals Variables: income, education, gender…. What the sample would be in the future Rules applied to every member in the sample Adventages: • Realistic data – Disadventages: • State transformations difficut to estimate • No agent-agent interaction – agent are isolated only interact with their environments • Early simulations in social science (1957)
CA v. s. ABM • • CA: interraction with their neighbor with simple rules CA agents have simple states usually a binary variable – alife – death, – not buy - buy, has the opinion – does not have • Dynamics of physical, chemical systems • E. g. : Game of life
6 ABM Applications • Eaarly adapting disiplines – chemistry, biology, material science • Second wave – – natural - physics, social – demography, political science, sociology geography - GIS crowd simulations • Latter – business, economics, . . .
Social Science Applications • Economics • Demogrphy • Political science – party competitions – voting behavior • • Socialogy / Antropology History Law Interdisiplinary – Science dynamics – soio-technical systems
Business/MIS • Business – – Finance Marketing / e-merketing Organizational behavior Operations management • Supply chain management / logistics – MIS • User modeling, value of information, e-business, e-auctions
Modeling Examples • Urban models -Schelling(1971, 1978) – Racial segregation – Grid cells, – Two types – rad, green • Opinion dynamics – Agents have opinions -1 to +1 and degree of doubt – Interact randomly • Consumer behavior • Marketing – – Viral marketing WOM effects Efficiency of marketing strategies Dynamics of markets: U-Mart project
Modeling Examples (cont. ) • Industrial networks – Links between firms – Inovation networks- biotechnology, ICT – Clustering of industries • Business ecosystems • Supply chain management – Effectiveness of management policy – Order fulfilment – Procter & Gamble
Business/MIS Examples • Diffusion – New product, technology, innovations • Markets – modeling software markets – versioning decisions timing of upgrading and how much and when • Financial merkets – Santa Fe Stock market – speculative behavior • Auctions – efficiency, profitability of e-auction mechanisms
Business/MIS Examples (cont. ) • Strategic management – Profitability, efficiencey of business strategies – Competitive or cooperative strategies – outsourcing • Organizational impact of information systems • Modeling simulation of business processes – Common with discrete event simulation but – ABMS enables including behavior of humans • Social Networks – Behaviour in social media – Dynamics off/on social networks • How social networks evolve over time • Network of networks
Business/MIS Examples (cont. ) • Industrial clusters – Similar firms in terms of what they produce (good services) – Tend to be locatyed in the same geographical regions • Software Engineering – Software upgrade quality improvement decisions in prsense of network effects • Modeling competition considering product life cycle diffusion of influences
Decision Support Systems (DSS) • ABMs can be embedded into DSS to perform – What if analysis – Sensitivity analysis – Senario analysis • User interface • Model base – – OR - optimzation – linear programming Statistical Analytical Simulation: ABM, SD, DES
Example: Simple Population Dynamics • How population of a country/region evolves over time • Assumption: Population of a country increrases proportional with the current value of its population • SD – one variable representing population N(t) as a function of time – homogenous • d. N/dt = g*N – rate of change of population is proportional to curent value of N • g: yearly growth rate of population • first order homogenous differential equation
Analytical Solution • Analytical solution even with frashman calculus d. N/N = gdt integrating both sides In. N + C = gt initial condition at time t=0 N= N 0, In. N + C = g*0 so C = - In. N 0, In. N – In. N 0 = gt In. N/N 0 = gt taking exponent of both sides N/N 0 = egt, N = N 0 egt,
As an emprical model • N 0 : the popution at an arbitary time calssed zero • g: yearly growth rete to be estimated from real population data • time(years) population(millions) 1970 35 1975 39 1980 42
Simulation in SD • The differential equation can be simulated as well • Excel simulation • given an initial population and a estimated g value • project population over time
ABM model • At time 0 • create set of agents representing age, gender, education, income, etnicity, geography of population • Each agent has a type has different fertality rate • As time progress – – with a probability have a chiild may die or migrate to another country new agents may migrate to the country but deterministically age increses by say 1 year
Example: Predator-Prey Interractions • Lotka-Volterra differential equations d. Pred/dt = K 1*Prad*Prey – M*Pred d. Prey/dt = B* Prey - K 2*Prad*Prey Two coupled nonlinear diferential equations ABM State mehanisms They have enery İncreass when eat decreases when move Prey may eat grass Predators eat prey
7 Resources • Associations: – North Americal Assoc. for Computational and Organizational Sciences – Posific Asean Assoc. for Agent-Based Approaches in Social Systems Science – Eurapean Socaal Simulation Assoc. • Journal: – Journal of Artifical Societies and Social Simulation • web sides: – Acent Based Computational Economics by Tesfatsion • Handbook of Computational Economics Vol 2 – by Judd and Tesfation
Books • Gilbert, N. , Agent-Baded Models, Saga Pubnlications, 2008. • North N. , J. , Macal, C. M. , Managing Business Compoexity: Discovering Strategic Solutions with Agent. Based Modeling and Simulation, Oxford University Press, 2008. • Railsback, S. , F. , Grimm, V. , Agent-Based and Individual. Baded Modeling: A Practical Introduction, Princeton University Press, 2011. • Robertson, D. , A. , Caldart, A. , . The Dynamics of Strategy: Mastering Strategic Landscapes of the Firm, Oxford University Press, 2009. –
8 Conclusion • Simulation in social science – third way of doing research • ABMS – buttom up – agnets • heterogenous • adaptive, learning behavior – interractions – emergence
- Slides: 57