Agent-based models of animal behaviour From micro to macro and back.

Transcription

1 Agent-based models of animal behaviour From micro to macro and back. Ellen Evers, Behavioural Biology, University Utrecht.

2 Outline... - genome cell organism group population ecosystem -... Organization, structure, patterns (spatial, temporal, social,...) Behaviour of animal groups (coordination) Self-organization (complexity arises from simple rules) Tool to understand and study: Agent-based Models (ABM)

3 Flocking Examples real: Starlings, ants (movie), fish

4 Collective motion Bird flocks (starlings), insect swarms (flies), mammal herds (wildebeest), fish schools (herring)

5 Fish schooling School level: Locomotion: Simultaneous change of direction Defense: Flash expansion, fountain effect, aggregation No disorganized crowd, but coordinated behaviour between members of the group! HOW to get structure, patterns, organization in a system?

6 Organization Leader: Well informed (top-down) Provides everyone with instructions No! Fish change position within school Blueprint, Recipe, Template: Plan or procedure Fixed, not very flexible No! Unlikely in big groups

7 Fish schooling Individual level: Vision (location of neighbours) Lateral line (orientation of neighbours) Coordination only with nearest neighbours, whole swarm locomotion through: SELF-ORGANIZATION!

8 Self-Organization... process in which a pattern at the global level of a system emerges solely from numerous interactions among the lower-level components of the system. (Camezine et al, 2001) No external, centralized control (no leader) Local information, simple rules (no global plan, overview) ORGANIZATION! (bottom-up) FISH SCHOOLING? COLLECTIVE MOTION?

9 Fish schooling Self-Organization Hypothesis: Few simple rules in response to local information from neighbouring fish generate schooling patterns without directly coding for them. (Huth & Wissel, 1992) Schooling rules: 1. Cohesion 2. Separation 3. Alignment

10 Fish schooling Self-Organization Hypothesis: Few simple rules in response to local information from neighbouring fish generate schooling patterns without directly coding for them. (Huth & Wissel, 1992) To test hypothesis and understand mechanism: Models?! Agent-Based Models (ABM)

11 Models What is a model?

12 Models

13 Models

14 Models

15 Models

16 Models Definition: Representation of a system of entities, phenomena, or processes to simplify, visualize, simulate, manipulate and gain intuition about the entity, phenomenon or process being represented. Simplified Realistic Understandable Predictive Explaining Precise Simplification versus Realism

17 Collective motion Model rules: 1. Cohesion 2. Separation 3. Alignment Craig Reynolds 1987: bird flocking (BOIDS) Huth & Wissel 1992: fish schooling Hooper, 1999: predator avoidance (cool school)

18 Collective motion Conclusions from the model: Simple rules can account for complex pattern No leader, no external cue nessecary Schools could be self-organizing No proof of real mechanism Proven: possible explanation works

19 2nd part

20 Scientific models Spatial CA nonlinear dynamical ordinary AGENT-BASED differential equations MODELS deterministic stochastic (ABM) IBM monte carlo simulation IOM partial differential equations agent-based models complex systems ORDINARY static probabalistic ABM (ODE) DIFFERENTIAL EQUATIONS difference equation ODE PDE cellular automata linear

21 Predator-prey system + sheep - + wolves - a) data: b) model: sheep sheep wolves wolves

22 - + sheep wolves - +

23 - + sheep wolves - + ODE: ABM: Population densities Discrete indiviuals sheep wolves

24 - + sheep wolves - + ODE: ABM: Equations Rules (if-then, loop, scheduling) ds/dt dw/dt = bs psw = psw - dw Sheep = + birth*sheep pred.*sheep*wolves Wolves = + pred.*sheep*wolves death*wolves SHEEP: If I meet wolve: die! With chance = birthrate: Reproduce! WOLVES: If I meet sheep: energy +1 If energy (from sheep) = 0: die! With chance = birthrate: Reproduce!

25 - + sheep wolves - + ODE: ABM: Deterministic Stochasticity (pseudo-random numbers) sheep wolves... to represent variability of processes that cannot or need not to be modelled deterministically After Grimm 2005

26 - + sheep wolves - + ODE: ABM: Non-spatial Homogenity Well mixed Explicitly spatial Heterogenity Local interactions (bottom-up, adaptive) Patterns

27 - + sheep wolves - + ODE: ABM: Homogeneous population Individual variation (age, colour, knowledge, spatial position, size, sex,...)

28 Agent-based Models = Agent-based Models (ABM) = Individual-based Models (IBM) = Individual-oriented Models (IOM)... describe behaviour, variability and interactions of autonomous individuals. After Grimm 2005

29 Collective foraging Ants: Trail formation, transporter recruitment, efficient path finding, PHEROMONES!

30 Collective foraging Collective decision of colony: Selection of richest of two foods Selection of shortest path Hyp 1: directed by leader (queen?) Hyp 2: Self-organization Global pattern emerges from simple rules and local information

31 Collective foraging Foraging rules: Execute random walk or follow strongest pheromone trail Food found: walk home & leave pheromone trace Resnick, M. (1994)

32 Collective foraging Self-organization: Simple rules No global knowledge, local information Self-reinforcing phermone trails Feedback via environment Emergence of accurate (not perfect!) collective decision: efficient path selection.

33 Self-organization Mechanism: Information from neighbours Information from local environment (stigmergy) Feedback from emerging structure (control, signal)

34 Emergent properties... arise out of interactions between lower-level entities, yet are novel and irreducible to them. (Stanford Encyclopedia of Philosophy) Emergent property: Order, organization, pattern on group level Entities: Individuals without global knowledge, plan or intent

35 Emergence Self-organization Complex Systems physical and chemical systems (liquid) biological systems (cell, brain) social systems and organizations (stock market, social structures) Complex system patterns often emerge through self-organization.

36 Complex Systems Multiple levels of organization! Individual < subgroup < population Interconnected, interdependent Higher level affects lower one Higher level has behaviour on its own Emergent property: Substrate to evolution

37 3rd part

38 ABM in science? Criticism: Too complex to be understood Impossible to include everything in a model Too many parameters unknown Hard to test Parameters can tweak everything Gaming After Grimm 2005

39 Modelling cycle After Grimm 2005

40 Modelling cycle Formulate the question: Filter for essential processes Not model per se, but for specific purpose After Grimm 2005

41 Modelling cycle Assemble hypotheses: Starting point: conceptual model (drawing) Based on empirical data, experience, theory, gut feeling What do we expect to see? PATTERNS (at multiple levels) After Grimm 2005

42 Modelling cycle Chose model structure: Which entities? Which processes? How to model/represent processes? After Grimm 2005

43 Modelling cycle Implement model: Write program code (if-then rules, loops) After Grimm 2005

44 Modelling cycle Analyze the model I: Controlled simulation experiments (visual debugging, process understanding, extreme tests) Design and analysis just as with real experiments (statistics on program generated data) After Grimm 2005

45 Modelling cycle Analyze the model II: Much freedom and control in experimental setup (upscale numbers and time, hypothetical thought experiments) Can the model reproduce PATTERNS? (at multiple levels!!!) After Grimm 2005

46 Modelling cycle Analyze the model IIIa: Change parameter/rule: Group Group Effect on system behaviour? Same parameter setting: (knockout, sufficient & neccessary) Individual Individual Group Individual Group Individual Changing parameter setting: Group Individual Group Individual Group Individual Group Individual After Grimm 2005

47 Modelling cycle Analyze the model IIIb: Multiple runs, same parameter set (account for stochasticity) Same parameter setting: Group Group Group Group Individual Individual Individual Individual After Grimm 2005

48 Modelling cycle Communicate the model: Materials & Methods Reproducability of results After Grimm 2005

49 ABM a scientific method Model validation: New hypotheses from the model Not built into the model beforehand Can be tested in real system After Grimm 2005

50 Primate social cognition Social structure: dominance hierachy Spatial structure: centrality of dominants

51 Primate social cognition DOMWORLD (Hemelrijk 1999, Hogeweg1988) Simple rules > complex behaviour 1. Grouping rules: Max view Near view Personal space

52 Primate social cognition Dom-value: rank (initially all the same) 2. Interaction rules: Prediction (mental simulation) Dominance contest: Win chance ~ rank Loser flees, winner chases Dom-value update according to expected outcome (winner-loser effect)

53 Primate social cognition Dominance hierarchy (differentiation from initially equal individuals) (individual variation!!!) Emergence of spatial structure (centrality of dominant individuals as unintentional side-effect) Hemelrijk, 1998 Stabilizing feedback from spatial structure on social structure: neighbors of similar rank no unexpected outcomes no big updates of dom-value

54 Primate social cognition INSIGHTS: No centripetal instinct necessary for spatial structure From side effect to evolutionary substrate Emerging spatial structure feeds back on social structure ABM internship:

55 Evolution in silico Evolution ingredients: Mutations: Inheritence of traits (to offspring) Chance of changing traits Selection: Define fitness measure Explicit selection (once in a while: take out 100 best and reproduce) Reproduction/death (dependent on fitness measure)

56 Conclusions Biological systems: Organisation / patterns Several possibilities: Central, fixed control (leader, blueprint,...) Self-organisation Interactions between individuals Emergence of pattern on group level

57 Conclusions ABM: useful tool to study / understand Multiple connected levels (individual, group) Behavioural rules Explicit spatiality: Locality Heterogenous space Individual variation (hierarchy, evolution) Heraklit: You never enter the same river twice.

58 Conclusions Models are useful: Development: Explicit formulation of processes Simplification: Identify essential (sufficient & nessecary) processes Scientific research tool: Controlled experiments (knockout, change parameter/rule) Series of runs (variation) Statistics on generated data Material & methods Reproduced results: validation Group Group Group Individual Individual Individual

59 Conclusions Models advantages: Freedom in experimental setup Little time, money, animals, facilities needed No ethical objections Generate new hypotheses Reveal general and unexpected properties of a system Help understanding a system Change way of thinking about a system Proust: The real voyage of discovery lies not in finding new landscapes, but in having new eyes.

60 Sources Modelling, general J. J. Bryson, Y. Ando, and H. Lehmann, Agent-based modelling as scientific method: a case study analysing primate social behaviour, Philosophical Transactions of the Royal Society B: Biological Sciences 362, no (2007): D. L. DeAngelis and W. M. Mooij, Individual-based modeling of ecological and evolutionary processes. Annual Reviews in Ecology, Evolution and Systematics 36: V. Grimm, Ten years of individual-based modelling in ecology: what have we learned and what could we learn in the future?, Ecological Modelling 115, no. 2 (1999): V. Grimm and S. F. Railsback, Individual-based Modeling and Ecology (Princeton University Press, 2005). V. Grimm et al., Pattern-Oriented Modeling of Agent-Based Complex Systems: Lessons from Ecology, Science 310, no (November 11, 2005): J. W. Haefner, Modeling Biological Systems: Principles and Applications, 2nd ed. (Springer, 2005). C. K. Hemelrijk, Understanding Social Behaviour with the Help of Complexity Science (Invited Article), Ethology 108, no. 8 (2002): P. Hogeweg, MIRROR beyond MIRROR, Puddles of LIFE, Artificial Life (1989): T. J. Prescott, J. J. Bryson, and A. K. Seth, Introduction. Modelling natural action selection, Philosophical Transactions of the Royal Society B: Biological Sciences 362, no (2007): Complexity, Emergence and Self-organization, general Biological examples of emergence and self-organization J. Buck, Synchronous Rhythmic Flashing of Fireflies. II., The Quarterly Review of Biology 63, no. 3 (1988): 265. S. Camazine et al., Self-Organization in Biological Systems: (Princeton University Press, 2003).

61 Sources Agent-based models of animal behaviour R. Axelrod, The Complexity of Cooperation: Agent-Based Models of Competition and Collaboration (Princeton University Press, 1997). R. Axelrod and W. D. Hamilton, The evolution of cooperation, Science 211, no (1981): H. de Vries and J. C. Biesmeijer, Modelling collective foraging by means of individual behaviour rules in honey-bees, Behavioral Ecology and Sociobiology 44, no. 2 (1998): H. de Vries and J. C. Biesmeijer, Self-organization in collective honeybee foraging: emergence of symmetry breaking, cross inhibition and equal harvest-rate distribution, Behavioral Ecology and Sociobiology 51, no. 6 (2002): C. K. Hemelrijk, Towards the integration of social dominance and spatial structure, Animal Behaviour 59, no. 5 (May 2000): , doi: /anbe C. K. Hemelrijk, Social phenomena emerging by self-organisation in a competitive, virtual world ( DomWorld ), 1st International Conference of the European Society for Social Simulation (ESSA 2003), Groningen, September (2003). P. Hogeweg, MIRROR beyond MIRROR, Puddles of LIFE, Artificial Life (1989): P. Hogeweg and B. Hesper, The ontogeny of the interaction structure in bumble bee colonies: A MIRROR model, Behavioral Ecology and Sociobiology 12, no. 4 (1983): C. Honk and P. Hogeweg, The ontogeny of the social structure in a captive Bombus terrestris colony, Behavioral Ecology and Sociobiology 9, no. 2 (1981): H. Lehmann, J. Wang, and J. J. Bryson, Tolerance and Sexual Attraction in Despotic Societies: A Replication and Analysis of Hemelrijk (2002), Modelling Natural Action Selecton: Proceedings of an International Workshop, Edinburgh. The Society for the Study of Artificial Intelligence and the Simulation of Behaviour (AISB) (2005). S. C. Pratt et al., An agent-based model of collective nest choice by the ant Temnothorax albipennis, Animal Behaviour 70, no. 5 (2005): M. Resnick, Turtles, Termites, and Traffic Jams: Explorations in Massively Parallel Microworlds (MIT Press, 1994). C. W. Reynolds, Flocks, Herds, and Schools: A Distributed Behavioral Model, Computer Graphics 21, no. 4 (1987): W. I. Sellers, R. A. Hill, and B. S. Logan, An agent-based model of group decision making in baboons, Philosophical Transactions of the Royal Society B: Biological Sciences 362, no (2007): D. J. van der Post and P. Hogeweg, Resource distributions and diet development by trial-and-error learning, Behavioral Ecology and Sociobiology 61, no. 1 (2006): D. J. van der Post and P. Hogeweg, Diet traditions and cumulative cultural processes as side-effects of grouping, Animal Behaviour 75, no. 1 (2008):