UNCLASSIFIED 1 Verification, Validation and Accreditation of Agent- Based Simulations Deborah Duong.

Slides:



Advertisements
Similar presentations
Introduction to Psychology
Advertisements

CHAPTER 1 WHAT IS RESEARCH?.
Situation Calculus for Action Descriptions We talked about STRIPS representations for actions. Another common representation is called the Situation Calculus.
Overarching Goal: Understand that computer models require the merging of mathematics and science. 1.Understand how computational reasoning can be infused.
Modeling and simulation of systems Slovak University of Technology Faculty of Material Science and Technology in Trnava.
Software Engineering Experimentation Experimental Terms Jeff Offutt
The Modeling Process Esmaeil Khedmati Morasae Center for Community-Based Participatory Research in Health Tehran University of Medical Sciences Winter.
Gautam Sanka. Analyze and Elucidate the behavior of complex systems Complex Systems Collection of interconnected elements (system) Behavior and Characteristics.
Participatory Simulation & Emergent Behavior Author : Uri Wilensky Presenter : Krunal Doshi.
FIN 685: Risk Management Topic 5: Simulation Larry Schrenk, Instructor.
CPSC 411, Fall 2008: Set 12 1 CPSC 411 Design and Analysis of Algorithms Set 12: Undecidability Prof. Jennifer Welch Fall 2008.
1 Undecidability Andreas Klappenecker [based on slides by Prof. Welch]
Doing Social Psychology Research
Data Mining.
Scientific Thinking - 1 A. It is not what the man of science believes that distinguishes him, but how and why he believes it. B. A hypothesis is scientific.
1 Validation and Verification of Simulation Models.
Overview of Software Requirements
Developing Ideas for Research and Evaluating Theories of Behavior
Scientific method - 1 Scientific method is a body of techniques for investigating phenomena and acquiring new knowledge, as well as for correcting and.
How Psychologists Ask and Answer Questions
Science Inquiry Minds-on Hands-on.
Bottom-Up Coordination in the El Farol Game: an agent-based model Shu-Heng Chen, Umberto Gostoli.
Thinking Like a Modern Economist 6 Economics is what economists do. — Jacob Viner CHAPTER 6 Copyright © 2010 by the McGraw-Hill Companies, Inc. All rights.
Research method2 Dr Majed El- Farra 1 Research methods Second meeting.
Research and Statistics AP Psychology. Questions: ► Why do scientists conduct research?  answer answer.
Profile and a quick introduction Software Engineering: ) هندسة البرمجيات (in Arabic: is the branch of computer science Designed to develop a set rules.
UNCLASSIFIED 1 Top 3 M&S Challenges for SSTR Operations Developing Epistemologically Sound Standards for Analysis Developing Adequate Technology to Represent.
1 Performance Evaluation of Computer Networks: Part II Objectives r Simulation Modeling r Classification of Simulation Modeling r Discrete-Event Simulation.
Scatterplots, Associations, and Correlation
TEA Science Workshop #3 October 1, 2012 Kim Lott Utah State University.
Evaluation of software engineering. Software engineering research : Research in SE aims to achieve two main goals: 1) To increase the knowledge about.
1 OM2, Supplementary Ch. D Simulation ©2010 Cengage Learning. All Rights Reserved. May not be scanned, copied or duplicated, or posted to a publicly accessible.
Big Idea 1: The Practice of Science Description A: Scientific inquiry is a multifaceted activity; the processes of science include the formulation of scientifically.
Types of Research (Quantitative and Qualitative) RCS /11/05.
Chapter 1: Psychology, Research, and You Pages 2 – 21.
Art 315 Lecture 6 Dr. J. Parker. Variables Variables are one of a few key concepts in programming that must be understood. Many engineering/cs students.
The Nature of Modeling and Modeling Nature. “The sciences do not try to explain, they hardly even try to interpret, they mainly make models… The justification.
1 Lesson 8: Basic Monte Carlo integration We begin the 2 nd phase of our course: Study of general mathematics of MC We begin the 2 nd phase of our course:
URBDP 591 I Lecture 3: Research Process Objectives What are the major steps in the research process? What is an operational definition of variables? What.
Hypothesis & Research Questions Understanding Differences between qualitative and quantitative approaches.
Conceptual Modelling and Hypothesis Formation Research Methods CPE 401 / 6002 / 6003 Professor Will Zimmerman.
© 2008 McGraw-Hill Higher Education The Statistical Imagination Chapter 1. The Statistical Imagination.
Research Methods in Computational Informatics IST 501 Fall 2014 Dongwon Lee, Ph.D.
Simulation is the process of studying the behavior of a real system by using a model that replicates the behavior of the system under different scenarios.
Gile Sampling1 Sampling. Fundamental principles. Daniel Gile
Memory Management. Memory  Commemoration or Remembrance.
Hazard Identification
1 The Theoretical Framework. A theoretical framework is similar to the frame of the house. Just as the foundation supports a house, a theoretical framework.
Introduction to Earth Science Section 2 Section 2: Science as a Process Preview Key Ideas Behavior of Natural Systems Scientific Methods Scientific Measurements.
Scientific Methods and Terminology. Scientific methods are The most reliable means to ensure that experiments produce reliable information in response.
The Representation of Uncertainty for Validation and Analysis of Social Simulations TRADOC Analysis Center – Monterey 21 September 2010 Unclassified Debbie.
Sampling and estimation Petter Mostad
Research for Nurses: Methods and Interpretation Chapter 1 What is research? What is nursing research? What are the goals of Nursing research?
Computer Simulation of Networks ECE/CSC 777: Telecommunications Network Design Fall, 2013, Rudra Dutta.
RULES Patty Nordstrom Hien Nguyen. "Cognitive Skills are Realized by Production Rules"
Chapter 1 What is Biology? 1.1 Science and the Natural World.
WHAT IS RESEARCH? According to Redman and Morry,
…empowering communities through modeling and adaptive management Sustaining Ecological Communities Through Citizen Science and Online Collaboration.
Copyright © 2015 Inter-American Development Bank. This work is licensed under a Creative Commons IGO 3.0 Attribution-Non Commercial-No Derivatives (CC-IGO.
Conducting Research Psychology, like chemistry and biology, is an experimental science, assumptions must be supported by scientific evidence. It is not.
CS223: Software Engineering Lecture 25: Software Testing.
The Scientific Method. Scientifically Solving a Problem Observe Define a Problem Review the Literature Observe some More Develop a Theoretical Framework.
Modelling & Simulation of Semiconductor Devices Lecture 1 & 2 Introduction to Modelling & Simulation.
Section 2: Science as a Process
Object oriented system development life cycle
Future Studies (Futurology)
Conceptual Frameworks, Models, and Theories
Introduction Artificial Intelligent.
CSCE 411 Design and Analysis of Algorithms
Principles of Science and Systems
Presentation transcript:

UNCLASSIFIED 1 Verification, Validation and Accreditation of Agent- Based Simulations Deborah Duong

UNCLASSIFIED 2 Purpose To introduce Agent-Based Simulation To propose measures of effectiveness for Agent-Based Simulation

UNCLASSIFIED 3 What is an Agent-Based Simulation? “Agent-Based Simulation” (ABS) is broadly defined –An ABS is a simulation in which entities have “agency” –Agents can perceive and behave in their environment based on goals Agent-Based Simulation is used for modeling living systems –Biological and social systems –Non-living systems are mindless, and therefore don’t have “agency” The concept of “emergence” is important –Agents behave according to one set of rules –New patterns “emerge” from individual behaviors –Emergence is micro-macro integration

UNCLASSIFIED 4 How does Agent-Based Simulation Compare ? Other methods that don’t involve agency or minds are also used to describe living systems –Discrete Event Simulation Events of a process are scheduled to occur at discrete points –System Dynamics Simulation Looks at the flow of “fluid” levels over time Time delays are important –Social Networks Patterns in the arrangement of entities to each other are important These methods are at their best when modeling “non-mental” phenomena –Ecology Predator-Prey cycles –The Economy Cycles not based on “beliefs” (like the stock market is) –Any time entities act similarly Everybody eats! Non-agent simulation methods model flows and arrangements of “averaged” entities –Their “State” does not change, because entities are not modeled explicitly –They are not “networked” –They are viewed from an external, “etic” standpoint

UNCLASSIFIED 5 Why some Computational Social Scientists prefer ABS –Their preference depends on their feelings on the importance of “agency” and minds –They may believe that other tools are not as rich Other tools tend to make “heroic assumptions” They often can not model the crux of the problem They are more descriptive than causal –North and Macal: “We believe that in the future virtually all computer simulations will be agent-based because of the naturalness of the agent representation and the close similarity of agent models to the predominant computational paradigm of object-oriented programming.”

UNCLASSIFIED 6 Cognitive vs. Reactive Agents Agent Types CognitiveReactive Meaning changesMeaning is Hard Coded Interpretations come from Autonomous Perception Interpretations come from Copying other Agents Learn based on ExperiencesReact the same way every time Coevolves: behavior changes social structure while social structure changes behavior New starting conditions form different patterns but rules of behavior do not change during the simulation Heavy ComputationLight Computation Typically uses Machine Learning TechniquesMay use static rules

UNCLASSIFIED 7 Data-Based vs. Theory-Based ABS Agent Based Simulation Types Data-BasedTheory-Based Concerned with modeling a single instance of what actually happened and will happen Concerned with modeling what is possible, based on theoretical principles Is initialized with the detailed data of a scenario Starts with a random “primordial soup” from which data emerges Purpose is to explore plausible next states given the initial state Purpose is to model causes of states Stopping start: the initial state is not necessarily something that could emerge from the simulation itself Running start: Difficult to match to a particular data set: data must be “grown” from a previous state More descriptive: to fit data, correlations tend to be enforced without the modeling of cause More causal: No data to fit, only relations between events

UNCLASSIFIED 8 Agent Based Simulation and VV&A Verification –Determination of whether a simulation expresses a theory well Validation –Determination of whether a simulation has fidelity with the real world Accreditation –Determination that a simulation is useful for analysis of a particular domain Verification, Validation and Accreditation of agent based models is problematic –VV&A originated in physics models –The nature of social science has implications for agent based VV&A

UNCLASSIFIED 9 Agent-Based Simulation and Verification The more a simulation has the power to express a theory, the more the simulation is verified –A System Dynamics model of a verbal theory wouldn’t have a high degree of “verification” unless that theory was about time-delays The referent of any mathematical or simulation model is a theory –In physics based models, verification is “doable” In physics-based models, verification is mainly about bugs Replication, or using a different method to simulate the same theory, can help debug agent based social models In social-science based agent models, verification is the central issue –Verification is about technology to represent an idea Newton had the technology of the calculus –The technology to simulate social theories is not trivial For example, a social theory about human learning may need a computer that can match a human in learning –With knowledge of available tools and creativity, Verification is just a matter of good (scientific) taste, for now

UNCLASSIFIED 10 The Social Literature as the Referent Fitting raw data is not enough for verification –Data can be over-fitted –One could “simulate” by never addressing cause, by only making correlated things appear magically Since “why” is not modeled, the simulation is not generally applicable If it wont model a new situation, it wont model itself well either –If there are no causes a level under the phenomena you model, you are only describing, not analyzing You can not explore the new levers to change outcomes, other than the ones you put in the simulation to begin with Data should be fitted through a theory of social science –Thoughtful models in the social literature are preferred to models from other fields Just because we have the tools to describe time delays, physical phenomena, and epidemiology doesn’t relate them to social theory Knowledge of all tools is needed to model the richness of the social world Tested by: surveying the relative frequency of issues in the social literature and comparing to the relative frequency of issues in an ABS

UNCLASSIFIED 11 Agent-Based Simulation and Validation The more explanatory power an agent-based simulation has, the more the simulation is validated A simulation model should match the data in the world in the way that its theory matches it –Validation of agent based simulation is dependant on verification: If an agent based simulation is not first verified, it will not be valid –Validation of agent based simulation is dependant on the explanatory power of its referent theory as well –Technology that enables verification enables exploratory creation of theories with explanatory power

UNCLASSIFIED 12 What can we expect from an ABS? To address validation, let us ask, what can we expect from a theoretically perfect ABS? –Even if the agent based model was completely correct, it still could not do long term prediction The social world is full of “Schelling Points”: arbitrary phenomena –We can expect it to display similar patterns to the real world, but not the exact data of the real world It should have the same correlative patterns –Links between events in a simulation should have a similar strength to links between corresponding events in the real world It should develop a distribution of plausible results similar to the real world –Tested by “separating the test set from the training set” It should be able to make a short term prediction of “types” of phenomena –A live connection to data is essential An agent based simulation *is* a theory –It is a theory represented in a form amenable to computation –The theory that best matches the (patterns in) data is the best theory

UNCLASSIFIED 13 Validating Agent-Based Simulations –Data-Based vs. Theory-Based Agent models: How do we simulate both theory and data well? The trajectory of a theory-based simulation can be made to pass through particular data –Random number massaging –Co-evolutionary “seeding” Because the data emerges from the simulation itself, it models the next state better –It is validated if it models not only patterns in data, and the social literature well, but it also models causation well Ockham’s razor: If many known phenomena emerge from a few known phenomena, you have modeled a cause well

UNCLASSIFIED 14 Agent-Based Simulation and Accreditation Rating for a usage in a domain is based on correctness of past usage in that domain –Pattern-based correctness Social Science simulations are so complex, that scientific insight is needed in each new application –There is no way to generalize what tool will always be good in advance for what domain –Accreditation efforts should be devoted to confirming that a simulation does have expressive and explanatory power after the tool is chosen for the application When is a model ready for use in analysis? –When it predicts patterns in data and the occurrence of “types” of events consistently when given new data

UNCLASSIFIED 15 Myths of Agent-Based VV&A “Chaos theory says there is no order, and any small change makes a big change in the outcome” –The social world is full of order and homeostasis “The cause of emergent phenomena is so complex that it is unknowable” –Cause is knowable because it is contained “in the box” –Scientific experiments can tease out cause Computer experiments can “hold all else the same” better than real world experiments can Statistics can find cause in Monte Carlo ABS

UNCLASSIFIED 16 Implications for Existing VV&A Techniques Exploratory Space and Risk Analysis –Testing simulations at the boundaries where it matters –Nonlinearities in agent-based simulation means we don’t know where it matters –“Agency” can be taken advantage of in strategic data farming Bottom-up VV&A –Making sure that the lower level is VV&A’d and that will take care of the upper level –But you don’t know what to emphasize in the lower level until after the emergence happens

UNCLASSIFIED 17 Summary Agent Based Simulations model “Agency” –ABS are best used when mental processes and dynamic networks are important ABS may be typed according to two dimensions –Cognitive/Reactive –Data-Based/Theory-Based There is hope for Agent Based Simulation Verification, Validation and Accreditation –We have ways to measure Similar patterns to the real world correlative data Match to the social theory in literature Explanatory power (Ockham's Razor)