Thomas French29th May 2013www.sandtable.comTechnical Challenges of Real-World Agent-Based ModellingThursday, 30 May 13
Outline• What is ABM?• Why use ABM?• Classic ABM example• Real World ABM• Three Key Technical ChallengesThursday, 30 May 13
“Essentially, all models arewrong, but some are useful”G.E. Box (1987)Thursday, 30 May 13
ABM in a nutshellAGENTENVIRONMENTSENSORSMESSAGESACTIONSPERCEPTSOBJECTACTUATORSBased&on&Bordini&et&al&&(2007)Thursday, 30 M...
Why are we talking about ABM?• It shows promise forunderstanding complexsystems:– heterogeneous andadaptive actors– comple...
Classic ABM: Schelling Segregation Model• Developed by Thomas Schelling in 1970s.• Study racial segregation of populations...
Source: Eric FisherThursday, 30 May 13
Schelling Segregation ModelThursday, 30 May 13
Schelling Segregation ModelThursday, 30 May 13
Schelling Behaviour TreeThursday, 30 May 13
Real World ABMThursday, 30 May 13
QuitSIM Behaviour Tree-QUIT SIM 2QS Tree in Colour CensorThu May 30 2013Thursday, 30 May 13
QuitSIM Behaviour TreeTake upsmoking?Never smoker = 1Become smokerAge, genderDo nothingCut downattempt length, route,age, ...
Technical ChallengesBUILD VALIDATE EXPERIMENTDesigning*andbuilding*modelsBuildingConfidence*in*ModelsConductingLarge?ScaleE...
Building ModelsBUILDVALIDATEEXPERIMENTBehavioural+DataSurveyDataAssumptionsIntuitionAnalyse BuildIndividual+Agent+Attribut...
Validation - Building ConfidenceVALIDATEEXPERIMENTDoes the implementedmodel reflect thereal-world system?Thursday, 30 May 13
Validation – Establishing CriteriaA framework for evaluating state of validity of modelsfor on-going monitoring.VALIDATEEX...
Validation - ExamplesRepresented in a formal logic• linear-time temporal logic with extensionsInternal:(s_Att.gender = f) ...
Validation –Solving Multi-Criteria ProblemsVALIDATEEXPERIMENTThursday, 30 May 13
Validation - WorkflowVALIDATEEXPERIMENTSelect&ModelSelect&TestsSelect&Reference&DataConfigure&Test&SuiteExecute&Replications...
Experimentation -Approaches• Empirical Calibration• Sensitivity Analysis• Scenario Exploration• Goal-Directed SearchEXPERI...
Experimentation –Exploring Parameter SpacesEXPERIMENTSmall LargeExplore Exhaustive+SearchSimple+Random+Sampling,+Latin+Hyp...
Experimentation -Handling NoiseEXPERIMENTThursday, 30 May 13
Experimentation –Handling Output DataEXPERIMENTThursday, 30 May 13
Experimentation –Platform ArchitectureEXPERIMENTCATALOGREST APIWORKFLOWSCENARIOSANALYSISVALIDATIONOPTIMISATIONSERVICESmong...
Experimentation -Managing WorkflowEXPERIMENTThursday, 30 May 13
Thursday, 30 May 13
“Nothing is built on stone;all is built on sand. But we mustbuild as if sand were stone.”J.L. BorgesThursday, 30 May 13
Thanks for listening!thomas@sandtable.comwww.sandtable.comThursday, 30 May 13
Further studyBook:• John Miller and Scott Page: Complex AdaptiveSystems: An Introduction to Computational Modelsof Social ...
Upcoming SlideShare
Loading in...5
×

Technical Challenges of Real-World Agent-Based Modelling

184

Published on

A presentation given to Data Science London in May 2013

Published in: Technology
0 Comments
0 Likes
Statistics
Notes
  • Be the first to comment

  • Be the first to like this

No Downloads
Views
Total Views
184
On Slideshare
0
From Embeds
0
Number of Embeds
0
Actions
Shares
0
Downloads
12
Comments
0
Likes
0
Embeds 0
No embeds

No notes for slide

Transcript of "Technical Challenges of Real-World Agent-Based Modelling"

  1. 1. Thomas French29th May 2013www.sandtable.comTechnical Challenges of Real-World Agent-Based ModellingThursday, 30 May 13
  2. 2. Outline• What is ABM?• Why use ABM?• Classic ABM example• Real World ABM• Three Key Technical ChallengesThursday, 30 May 13
  3. 3. “Essentially, all models arewrong, but some are useful”G.E. Box (1987)Thursday, 30 May 13
  4. 4. ABM in a nutshellAGENTENVIRONMENTSENSORSMESSAGESACTIONSPERCEPTSOBJECTACTUATORSBased&on&Bordini&et&al&&(2007)Thursday, 30 May 13
  5. 5. Why are we talking about ABM?• It shows promise forunderstanding complexsystems:– heterogeneous andadaptive actors– complex interactions:interdependencies;feedback loops– dynamic environment• It provides an accessiblemetaphor for modelling– modelling individuals• More and more data isavailable for our models- Finer levels ofgranularity• Computing power isavailable on-demand- Costs continue toreduceThursday, 30 May 13
  6. 6. Classic ABM: Schelling Segregation Model• Developed by Thomas Schelling in 1970s.• Study racial segregation of populations emergingfrom individual discriminatory behaviours.Thursday, 30 May 13
  7. 7. Source: Eric FisherThursday, 30 May 13
  8. 8. Schelling Segregation ModelThursday, 30 May 13
  9. 9. Schelling Segregation ModelThursday, 30 May 13
  10. 10. Schelling Behaviour TreeThursday, 30 May 13
  11. 11. Real World ABMThursday, 30 May 13
  12. 12. QuitSIM Behaviour Tree-QUIT SIM 2QS Tree in Colour CensorThu May 30 2013Thursday, 30 May 13
  13. 13. QuitSIM Behaviour TreeTake upsmoking?Never smoker = 1Become smokerAge, genderDo nothingCut downattempt length, route,age, dependencyConsumemedia / ingestexperienceSmokerSmoker = 1Never SmokerNever Smoker = 1Consumemedia / ingestexperienceGet support?Set support flagPlanned orUnplanned?Do somethingaboutsmoking?motivation, events,price, GP, social,pregnant, media,random8#2013Thursday, 30 May 13
  14. 14. Technical ChallengesBUILD VALIDATE EXPERIMENTDesigning*andbuilding*modelsBuildingConfidence*in*ModelsConductingLarge?ScaleExperimentsHARD VERY,*VERY*HARD VERY*HARDThursday, 30 May 13
  15. 15. Building ModelsBUILDVALIDATEEXPERIMENTBehavioural+DataSurveyDataAssumptionsIntuitionAnalyse BuildIndividual+Agent+AttributesBehaviour+TreeEnvironment(e.g.+Media)Representative+PopulationData+SourcesSimulationComponentsThursday, 30 May 13
  16. 16. Validation - Building ConfidenceVALIDATEEXPERIMENTDoes the implementedmodel reflect thereal-world system?Thursday, 30 May 13
  17. 17. Validation – Establishing CriteriaA framework for evaluating state of validity of modelsfor on-going monitoring.VALIDATEEXPERIMENTVALIDATIONINTERNALVALIDATIONEXTERNALVALIDATIONModel&implemented&correctlyBehaviours&predicted&make&sense&/&are&logicalModel&stands&up&to&comparison&with&external&dataThursday, 30 May 13
  18. 18. Validation - ExamplesRepresented in a formal logic• linear-time temporal logic with extensionsInternal:(s_Att.gender = f) => (G (s_Att.gender = f) )G (!((s_Att.smoker = 1) && (s_Att.takeUp = 1)))G (!((s_Att.smoker = 1) && (s_Att.age < 11)))External:n_MSE (s_Val1.prevalence, r_Val1.prevalence)n_MSE (s_Val2.quit_atts, r_Val2.quit_atts)VALIDATEEXPERIMENTThursday, 30 May 13
  19. 19. Validation –Solving Multi-Criteria ProblemsVALIDATEEXPERIMENTThursday, 30 May 13
  20. 20. Validation - WorkflowVALIDATEEXPERIMENTSelect&ModelSelect&TestsSelect&Reference&DataConfigure&Test&SuiteExecute&ReplicationsSummarise&Individual&TestsSummarise&Test&SuiteThursday, 30 May 13
  21. 21. Experimentation -Approaches• Empirical Calibration• Sensitivity Analysis• Scenario Exploration• Goal-Directed SearchEXPERIMENTThursday, 30 May 13
  22. 22. Experimentation –Exploring Parameter SpacesEXPERIMENTSmall LargeExplore Exhaustive+SearchSimple+Random+Sampling,+Latin+Hypercube+Samplinge.g.+7+vars,+10/100+values+=+1+Trillion+parameter+setsSeek Exhaustive+SearchNoisy,+MultiEObjective+Evolutionary+AlgorithmsParameter+SpaceSearch+TypeThursday, 30 May 13
  23. 23. Experimentation -Handling NoiseEXPERIMENTThursday, 30 May 13
  24. 24. Experimentation –Handling Output DataEXPERIMENTThursday, 30 May 13
  25. 25. Experimentation –Platform ArchitectureEXPERIMENTCATALOGREST APIWORKFLOWSCENARIOSANALYSISVALIDATIONOPTIMISATIONSERVICESmongoDBMANAGERWORKER 1PLATFORMRabbitMQMESSAGINGhttp://sandtable.comSandtable Simulation PlatformCLIENTsimulationanalysisvalidation12k23NS3Sandtable)Simulation)PlatformThursday, 30 May 13
  26. 26. Experimentation -Managing WorkflowEXPERIMENTThursday, 30 May 13
  27. 27. Thursday, 30 May 13
  28. 28. “Nothing is built on stone;all is built on sand. But we mustbuild as if sand were stone.”J.L. BorgesThursday, 30 May 13
  29. 29. Thanks for listening!thomas@sandtable.comwww.sandtable.comThursday, 30 May 13
  30. 30. Further studyBook:• John Miller and Scott Page: Complex AdaptiveSystems: An Introduction to Computational Modelsof Social Life (2007)Coursera:• Scott Page: Model Thinking• https://www.coursera.org/course/modelthinkingThursday, 30 May 13
  1. A particular slide catching your eye?

    Clipping is a handy way to collect important slides you want to go back to later.

×