Emergent Computing for Natural and Social Complex Systems

Transcription

1 Emergent Computing for Natural and Social Complex Systems Professor Cyrille Bertelle bertelle LITIS - Laboratory of Computer Sciences, Information Technologies and Systemic Le Havre University - FRANCE JICCSE 2006 Al-Balqa Applied University, Jordan December 5-7, 2006

2 Outline 1 Complexity & Self-Organization 2 Emergent Computing Cellular Automata Modelling (Bak s sand pile & Schelling s seggregation Individual-Based Modelling (Agents and Ants) 3 Contributions Computing-Based Modelling for Living Systems Economical and Social Modelling 4 Conclusion Cyrille Bertelle Emergent Computing 2/80

3 Contributors Le Havre University D. Olivier, M.A. Aziz-Alaoui, F. Guinand, A. Cardon, J. Colloc, V. Jay, J.-L. Ponty. PhD students contributions B. Adouobo, P. Tranouez, K. Khatatneh, H. Kadri-Dahmani, G. Prévost, A. Dutot, F. Kebair, L. Jaff, R. Ghnemat, K. Mahboub, M. Nabaa. Main collaborations G.H.E. Duchamp (Paris 13 Univ.), M. Obaidat (Monmouth Univ.), M. Cotsaftis (ECE, Paris), S. Oqeili, A. Sheta and Z. Odibat (BAU). Cyrille Bertelle Emergent Computing 3/80

4 Complexity & Self-Organization Section Outline Complexity within Natural Systems Self-Organization Understanding Complexity and Current World Description Ecosystems and Climate change Complex Systems Concepts Dissipative Structures Cyrille Bertelle Emergent Computing 4/80

5 Complexity within natural and artificial Systems System as generic description Physic, Biology, Computer Science or Human Science give many examples of systems where the global behavior is the result of interactions between homogeneous or heterogeneous entities. Cyrille Bertelle Emergent Computing 5/80

6 Social animal systems Interaction and self-organized society Cyrille Bertelle Emergent Computing 6/80

7 Bacteria colonies Interaction and self-organization after a external stimuli Cyrille Bertelle Emergent Computing 7/80

8 Complexity & Self-organization Some elements of self-organization understanding Interactions and system formation are essential They are the basis of local or global properties The interactions are expressed within a network in permanent evolution Cyrille Bertelle Emergent Computing 8/80

9 Complexity and current world description Current world complexity within New Technologies Advances in Computer Science Huge Data Bases High Performance Computing Advances in Information Technology Satellites Internet Cyrille Bertelle Emergent Computing 9/80

10 Complexity and current world description Current world complexity within New Technologies From Computer Networks to World-Wide Communications Informational dynamic fluxes Increasing expectations and needs for information From New Models to New Challenges Geopolitics & World-Wide Economy are: Complex Systems Cyrille Bertelle Emergent Computing 10/80

11 Complexity and current world description Complexity in Urban System Modelling Cities are Buildings and spaces (parks, streets, etc); But also people: individuals and organizations. Modelling Objectives To Study the interactions of all these city constituents. To have effective decision making concerning urban development Complex system modeling is needed Cyrille Bertelle Emergent Computing 11/80

12 Ecosystems complexity Natural ecosystems complexity: from interactions to self-organized systems as a whole Cyrille Bertelle Emergent Computing 12/80

13 Climatic change From fiction to geopolitic Fiction: "Day after tommorrow" movie. Computation on natural systems. Reality: Kyoto. Geopolitic and North-South equilibrium interact with climatic evolution. Complexity as a whole from natural, economic, social and politic interaction Cyrille Bertelle Emergent Computing 13/80

14 Complex Systems Concept Interactive entities Systems involve a number of interacting entities: Each entity have limited and partial knowledge; Local relationships between these entities. Cyrille Bertelle Emergent Computing 14/80

15 Complex Systems Concept Emergent System Properties System properties emerge from collective interactions of its constitutive entities; Emergent properties are not expressed by the limited set of atomic rules applied at local level on the constitutive entities. Cyrille Bertelle Emergent Computing 15/80

16 Complex Systems Concept Emergent System Properties Emergent properties cannot be traced by separately analyzing the attributes of constitutive entities "The system is more than the sum of its parts" Cyrille Bertelle Emergent Computing 16/80

17 Complex Systems Concept System feedback over its entities Bilateral feedback: The system which emerge from entities, constraints theses entities to form it. Complex adaptive behavior & Self-organization Cyrille Bertelle Emergent Computing 17/80

18 Complex Systems Concept System feedback over its entities Two kinds of feedback: Positive one which amplify the system formation Negative one which regulate the system formation Cyrille Bertelle Emergent Computing 18/80

19 Complex Systems Concept Complex systems are open Complex adaptive systems are open systems (dissipative structures): They are crossed by information fluxes (like ecosystems are crossed by energetic fluxes) Cyrille Bertelle Emergent Computing 19/80

20 Complex Systems Concept Complex systems are open Opening is an evolutive process: The system structures can change because of these fluxes The system exhibits periods of fluctuation or bifurcations (sudden jumps in system that lead to new structures) Cyrille Bertelle Emergent Computing 20/80

21 Complex Systems and Dissipative Structures Ilya Prigogine Theory Better understanding of living systems with extension of classical thermodynamic laws. Dissipative structures are involved by continuous energetic fluxes crossing the system. They maintain themselves in a "stable" state far from equilibrium, including multiple feedback loops (non linear equations) Dissipative structure are islands of order in a sea of disorder. Cyrille Bertelle Emergent Computing 21/80

22 Brain is a dissipative structure Learning is a self-organization process of the brain crossed by informational or emotional fluxes Cyrille Bertelle Emergent Computing 22/80

23 Emergent Computing for Self-Organized Systems Section Outline Cellular automata Sand Pile Model from Per Bak Seggregation Model from Thomas Schelling Individual-Based Modelling Integrative Approach: Agent-Based Systems Modelling Ant Systems Cyrille Bertelle Emergent Computing 23/80

24 Definition Definition Cellular automata Regular grid of cells, each of them is characterized by a finite number of states Discrete and global time Local rules of cell interaction toward global emergent behavior Application for this work Sand pile model Segregation model Cyrille Bertelle Emergent Computing 24/80

25 Sand Pile Model from Per Bak Self-Organized Criticality & Sand Pile Model Per Bak proposes a model called selforganized criticality Sand pile formation is a basic example: Grains are dropped to form a sand pile; Pile self-organizes to critical states; Cyrille Bertelle Emergent Computing 25/80

26 Sand Pile Model from Per Bak Self-Organized Criticality & Sand Pile Model Sand pile formation is a basic example: Addition of small grains can result in "avalanche" : cascade of sand downs the edges of the sand pile; The magnitude of avalanche can vary from one grain to catastrophic collapses. Sand pile formation history is composed of many critical self-organized states. Cyrille Bertelle Emergent Computing 26/80

27 Sand Pile Model from Per Bak Self-Organized Criticality & Sand Pile Model If Z (i, j) 4 then Z (i, j) Z (i, j) 4 Z (i 1, j) Z (i 1, j) + 1 Z (i + 1, j) Z (i + 1, j) + 1 Z (i, j 1) Z (i, j 1) + 1 Z (i, j + 1) Z (i, j + 1) + 1 Cyrille Bertelle Emergent Computing 27/80

28 Sand Pile Model from Per Bak Self-Organized Criticality & Sand Pile Model If Z (i, j) 4 then Z (i, j) Z (i, j) 4 Z (i 1, j) Z (i 1, j) + 1 Z (i + 1, j) Z (i + 1, j) + 1 Z (i, j 1) Z (i, j 1) + 1 Z (i, j + 1) Z (i, j + 1) + 1 Cyrille Bertelle Emergent Computing 28/80

29 Sand Pile Model from Per Bak Self-Organized Criticality & Sand Pile Model If Z (i, j) 4 then Z (i, j) Z (i, j) 4 Z (i 1, j) Z (i 1, j) + 1 Z (i + 1, j) Z (i + 1, j) + 1 Z (i, j 1) Z (i, j 1) + 1 Z (i, j + 1) Z (i, j + 1) + 1 Cyrille Bertelle Emergent Computing 29/80

30 Sand Pile Model from Per Bak Self-Organized Criticality & Sand Pile Model If Z (i, j) 4 then Z (i, j) Z (i, j) 4 Z (i 1, j) Z (i 1, j) + 1 Z (i + 1, j) Z (i + 1, j) + 1 Z (i, j 1) Z (i, j 1) + 1 Z (i, j + 1) Z (i, j + 1) + 1 Cyrille Bertelle Emergent Computing 30/80

31 Sand Pile Model from Per Bak Self-Organized Criticality & Sand Pile Model If Z (i, j) 4 then Z (i, j) Z (i, j) 4 Z (i 1, j) Z (i 1, j) + 1 Z (i + 1, j) Z (i + 1, j) + 1 Z (i, j 1) Z (i, j 1) + 1 Z (i, j + 1) Z (i, j + 1) + 1 Cyrille Bertelle Emergent Computing 31/80

32 Sand Pile Model from Per Bak Self-Organized Criticality & Sand Pile Model If Z (i, j) 4 then Z (i, j) Z (i, j) 4 Z (i 1, j) Z (i 1, j) + 1 Z (i + 1, j) Z (i + 1, j) + 1 Z (i, j 1) Z (i, j 1) + 1 Z (i, j + 1) Z (i, j + 1) + 1 Cyrille Bertelle Emergent Computing 32/80

33 Sand Pile Model from Per Bak Self-Organized Criticality & Sand Pile Model If Z (i, j) 4 then Z (i, j) Z (i, j) 4 Z (i 1, j) Z (i 1, j) + 1 Z (i + 1, j) Z (i + 1, j) + 1 Z (i, j 1) Z (i, j 1) + 1 Z (i, j + 1) Z (i, j + 1) + 1 Cyrille Bertelle Emergent Computing 33/80

34 Sand Pile Model from Per Bak Self-Organized Criticality & Sand Pile Model If Z (i, j) 4 then Z (i, j) Z (i, j) 4 Z (i 1, j) Z (i 1, j) + 1 Z (i + 1, j) Z (i + 1, j) + 1 Z (i, j 1) Z (i, j 1) + 1 Z (i, j + 1) Z (i, j + 1) + 1 Cyrille Bertelle Emergent Computing 34/80

35 Sand Pile Model from Per Bak Self-Organized Criticality & Sand Pile Model If Z (i, j) 4 then Z (i, j) Z (i, j) 4 Z (i 1, j) Z (i 1, j) + 1 Z (i + 1, j) Z (i + 1, j) + 1 Z (i, j 1) Z (i, j 1) + 1 Z (i, j + 1) Z (i, j + 1) + 1 Cyrille Bertelle Emergent Computing 35/80

36 Sand Pile Model from Per Bak Self-Organized Criticality & Sand Pile Model If Z (i, j) 4 then Z (i, j) Z (i, j) 4 Z (i 1, j) Z (i 1, j) + 1 Z (i + 1, j) Z (i + 1, j) + 1 Z (i, j 1) Z (i, j 1) + 1 Z (i, j + 1) Z (i, j + 1) + 1 Avalanche Magnitude: 8 involved cases Cyrille Bertelle Emergent Computing 36/80

37 Sand Pile Model from Per Bak Self-Organized Criticality & Sand Pile Model From sand pile, we deduce a law: big avalanches are rare and small frequent in exponential way: N(s) = s u Using logarithm representation of abscisse, we obtain a linear graph. Cyrille Bertelle Emergent Computing 37/80

38 Sand Pile Model from Per Bak Self-Organized Criticality & Sand Pile Model This law can describe many self-organized complex systems: Earthquakes: for 1000 earthquakes of magnitude 4 on Richter scale, there is only 100 of magnitude 5 and 10 of magnitude 6; Linguistic: for 1000 occurrences of "the" in an English text, only 100 occurrences of "I" and 10 of "say"; Urban systems: Big cities are rare and small frequent in exponential way. Cyrille Bertelle Emergent Computing 38/80

39 Segregation Model from Thomas Schelling Spatial Emergent Patterns & Schelling Model T. Schelling (Nobel Price in Economic Sciences - oct 2005) He contributes to enhance the understanding of conflict and cooperation about social institutions He proposes a simple model of spatial segregation which can lead to self-organized phenomena Cyrille Bertelle Emergent Computing 39/80

40 Segregation Model from Thomas Schelling Schelling Model Micro-level Rules Two categories of persons share a territory. Schelling proposes a rule-based model based on simplified spatial territory using chessboard. Stay if at least a third of neighbors of his category Move to random location, otherwise Cyrille Bertelle Emergent Computing 40/80

41 Segregation Model from Thomas Schelling Spatial Emergent Patterns & Schelling Model In following figures, Schelling s model computed on a cellular automaton; Two categories of people (red and green points - blue: free space); Self-organized pattern formations are observed; Cyrille Bertelle Emergent Computing 41/80

42 Individual-Based Modelling IBM-Introduction Cellular Automata deal with simple spatial configuration and rule-based processes; Individual-Based Modelling deal with individual population over an environment: Initialy proposed for population dynamic studies by biologists, allowing to care of individual aspects not managed by analytical models describing global population number evolution; More generally IBM is used for modelling and simulation purpose; Used in Artificial Intelligence for Distributed Problem Solving where Individual are the entities (not only person) on which the problem decomposition is made. Cyrille Bertelle Emergent Computing 42/80

43 Agent-Based System Modelling Multi-Agent System is an integrative approach for Individual-Based Modelling = Cf. Rawan Ghnemat Presentation Cyrille Bertelle Emergent Computing 43/80

44 Ant Systems When Computer Science Researchers learn from Ant Colonies Ant systems are reactive agents systems suitable to implement self-organization; Bio-Inspiration: We try to understand the bahavior of artificial ants, especially the mechanism which lead to the self-organizataion of social insects; We implement in a formal way some artificial ant systems: Mathematical formulation + distributed implementation; Many applications in engineering computation: graph oriented optimization problems, tasks allocation, clustering. Cyrille Bertelle Emergent Computing 44/80

45 Ant Self-Organization Self-Organization from social insects Natural ants are able to organize themselves on various aspects... here they collaborate to manage a broken way, making living bridge. Cyrille Bertelle Emergent Computing 45/80

46 Ant Self-Organization Ant Foraging Deneubourg experiment: Find the optimal way from nest to food source Use of pheromone when food is find, during the come back Cyrille Bertelle Emergent Computing 46/80

47 Ant Self-Organisation Ants cimetery clustering Cyrille Bertelle Emergent Computing 47/80

48 Ant Self-Organization Ants cimetery custering Ants form piles of corpses to clean nests Each has elementary action unknowing the whole situation but only local information No supervisor to lead the piles formation Emergence of the clustering not designed for itself but only as result of ants interaction. Cyrille Bertelle Emergent Computing 48/80

49 Ant Systems Engineering Applications Application to TSP Application to TSP - Traveling Salesman Problem Find the shortest cycle which links N interconnected towns within weigthed graph with only one pass in each town. Model to use Graph where nodes C i are towns and weighted edges D ij mean distance between towns T ij (t) are pheromon quantity leave by ants on the edge (i, j) Cyrille Bertelle Emergent Computing 49/80

50 Ant System applied to TSP Graph visualisation of the interconnected towns Cyrille Bertelle Emergent Computing 50/80

51 Ant System Algorithm for TSP Synthetic algorithm (1) When an ant is on the town i, it computes the probability to go to the non yet visited town j by the formula: ( Tij (t) ) ( ) α 1 β Pij k (t) = l J k (t) D ij (T il (t)) α ( 1 D il ) β if j J k (t) 0 if j J k (t) Cyrille Bertelle Emergent Computing 51/80

52 Ant System Algorithm for TSP Synthetic algorithm (2) In the previous formula: J k is the set of towns not yet visited by the ant k; The numerator means that: The more there are pheromone (T ij ), the more the probability P ij is; The less the distance (D ij ) is, the more the probability P ij is; α and β allow to control the relative importance of the 2 previous parts. The denominator (sum of all possible numerators) allows to compute a probability value. Cyrille Bertelle Emergent Computing 52/80

53 Ant System Algorithm for TSP Synthetic algorithm (3) When an ant has find a solution as a good cycle between towns, it deposits some pheromone on all of the edges of the cycle, inversely proportional to the length of the cycle (L k ): δt k ij (t) = { Q L k if (i, j) is a edge of the cycle 0 elsewhere Where Q is a constant parameter Cyrille Bertelle Emergent Computing 53/80

54 Ant System Algorithm for TSP Synthetic algorithm (4) On each edge (i, j), the pheromone quantity is update from step t to step t+1, by adding all the contribution of each ant to previous pheromone quantity: T ij (t + 1) = ρt ij (t) + m δtij k (t) k=1 Where ρ is an evaporation factor which allow that some first path/solution can be remplaced by better ones. Cyrille Bertelle Emergent Computing 54/80

55 Ant System Algorithm Java Applet Demo for Ant Foraging From Antoine Dutot (LITIS - Le Havre University) Simulation of Deneubourg s experiment: Graph representation Previous formula implementation Visualization and adaptive process demo by moving dynamically one node Cyrille Bertelle Emergent Computing 55/80

56 Ant Clustering Principle Ants and other social insects deal with cooperative way in distributed clustering, like cimetery (see previous slide) Aggregations of objects Then small clusters formation which act on social insects as feed-back (stigmergie) Then the clusters increase Cyrille Bertelle Emergent Computing 56/80

57 Ant Clustering Synthetic algorithm (1) The algorithm consists in making moving in random direction a great number of autonomous ants inside an environment of objects to agregate them in clusters. At each step, an ant is in one of the 3 situations: 1 The ant moves without carrying anything and meet no object the ant continue to move randomly Cyrille Bertelle Emergent Computing 57/80

58 Ant Clustering Synthetic algorithm (2) 2 The ant moves without carrying and meet an object. The ant can take this object to carry it. The probability of the ant take the object is: P p = ( k1 ) 2 k 1 + f f is a value corresponding of the number of objects perceived in the ant neighborhood k 1 is a treshold: if f k 1 then Pp is near 1 if f k 1 then Pp is near 0 Cyrille Bertelle Emergent Computing 58/80

59 Ant Clustering Synthetic algorithm (3) 3 The ant moves and carries an object. The ant can leave this object on the ground. The probability of the ant to leave the carried object is: P d = ( f ) 2 k 2 + f f is a value corresponding of the number of objects perceived in the ant neighborhood k 2 is another treshold: if f k 2 then P d is near 0 if f k2 then P d is near 1 Cyrille Bertelle Emergent Computing 59/80

60 Ant Clustering Implementation NetLogo(1) Many platforms have been proposed since some years for the development of agents-based programming. We focus here on NetLogo: Authored by Uri Wilensky and under continuous development at Northwestern University (USA), Institute on Complex Systems, Center for Connected Learning and Computer-Based Modeling Whole integrated interface allowing usage of predefined models (using specific language). Cyrille Bertelle Emergent Computing 60/80

61 Ant Clustering Implementation NetLogo(2) Programmable modelling environment for simulating natural and social phenomena. Authoring graphical environment allowing to create our own model with a specific language NetLogo Demo Ants: corresponds to ant foraging on an unlimited plan (no graph support). Multi source of foods that are dynamically consumed Termites: corresponds to distributing clustering (very similar to previous real ant clustering photos) Cyrille Bertelle Emergent Computing 61/80

62 Ant Application to Collective Robotic Swarm Robotic Principles Ants are nice natural model but... not efficient industrial workers! Principle: adapt the ant organizations and associated technics to collective/swarm robotics and make efficient similar algorithms for industrial purpose. Cyrille Bertelle Emergent Computing 62/80

63 Ant Application to Collective Robotic Why Swarm Robotic? Similar evolution than from Artificial Intelligence to Distributed Artificial Intelligence; Robotic can be classify in two parts: Human like robots: very sophisticated machine but which need powerful central control. The failure of the robot make failed the task to do. Collective robot system: coordination of many simple robots system with simple individual behavior but allowing the emergence of collective behavior like in social insects systems. One robot can fail and the system adapt without it. Cyrille Bertelle Emergent Computing 63/80

64 Ant Application to Collective Robotic Swarm Robotic: Example of applications Foraging can be adapted to ore exploitation in hard environment for human (on Mars planet for exemple) Foraging can be adapted to research and human rescue inside a tunnel under fire (hard condition for human and robutness of the solution, even some robots are dammaged) Clustering can be adapted to cooperative transport by swarm of robots Extension possible in swarm robotic: increase the efficience of each robot by learning process based on genetic algorithm Cyrille Bertelle Emergent Computing 64/80

65 Contributions Section Outline Computer-based Modelling for Living Systems Living Complex System & Distributed Computer Systems Dynamical Graphs and Networks Structures Detection and their management Dynamical Distribution Computation using Ant System Economical and Social Modelling Urban Car Traffic Genetic automata and spatial organizations for Decision Systems Support Emotional Aspect in Decision Makingf Cyrille Bertelle Emergent Computing 65/80

66 Computer-Based Modelling for Living Systems Living Complex Systems & Distributed Computer Systems 2 goals : Understanding and modelling the organization and self-organization of living systems using appropriate models and simulations, Take inspiration from living systems to design new methodologies, new models and new algorithms suited to parallel and distributed computer systems. Cyrille Bertelle Emergent Computing 66/80

67 Computer-Based Modelling for Living Systems Living Complex Systems & Distributed Computer Systems 2 goals : Understanding and modelling the organization and self-organization of living systems using appropriate models and simulations, Take inspiration from living systems to design new methodologies, new models and new algorithms suited to parallel and distributed computer systems. Cyrille Bertelle Emergent Computing 66/80

68 Computer-Based Modelling for Living Systems Living Complex Systems & Distributed Computer Systems 2 goals : Understanding and modelling the organization and self-organization of living systems using appropriate models and simulations, Take inspiration from living systems to design new methodologies, new models and new algorithms suited to parallel and distributed computer systems. Cyrille Bertelle Emergent Computing 66/80

69 Computer-Based Modelling for Living Systems Cyrille Bertelle Emergent Computing 66/80

70 Computer-Based Modelling for Living Systems Research Developments Modelling and simulating aquatic ecosystems (fluid flow and trophic chains organizations in multi-scale and multi-models approaches), Bioinformatic with partialy unknown data (genomic using ant systems) Dynamical distribution of Multi-agents simulation over computer Network, dealing with load balancing. Services execution and spreading over distributed systems composed of mobile ressources. Cyrille Bertelle Emergent Computing 66/80

71 Dynamical Graphs and Networks Motivations Interactive entities set (ecosystem) : Interaction graph Organizations detection (multi-scale description) and dynamicaly re-implementation of these organizations inside the simulation, using for example multi-models approaches (IBM and equational ones), Multi-agent simulation : Communication graph Communication minimization and load balancing, Ad hoc networks diffusion : Connexion graph Maximisation of the number of the most quickly reached stations and minimisation of the sending message number between them. Cyrille Bertelle Emergent Computing 67/80

72 Dynamical Graphs and Networks Motivations Interactive entities set (ecosystem) : Interaction graph Organizations detection (multi-scale description) and dynamicaly re-implementation of these organizations inside the simulation, using for example multi-models approaches (IBM and equational ones), Multi-agent simulation : Communication graph Communication minimization and load balancing, Ad hoc networks diffusion : Connexion graph Maximisation of the number of the most quickly reached stations and minimisation of the sending message number between them. Cyrille Bertelle Emergent Computing 67/80

73 Dynamical Graphs and Networks Motivations Interactive entities set (ecosystem) : Interaction graph Organizations detection (multi-scale description) and dynamicaly re-implementation of these organizations inside the simulation, using for example multi-models approaches (IBM and equational ones), Multi-agent simulation : Communication graph Communication minimization and load balancing, Ad hoc networks diffusion : Connexion graph Maximisation of the number of the most quickly reached stations and minimisation of the sending message number between them. Cyrille Bertelle Emergent Computing 67/80

74 Structure Detection and their management Principle et goal : Detection of the structures which appear during the simulation and managing this information for load balancing. Cyrille Bertelle Emergent Computing 68/80

75 Structure Detection and their management Structures and interactions : fusion Interaction of elements of the same category leads to fusion and agregation : Cyrille Bertelle Emergent Computing 69/80

76 Structure Detection and their management Structures and interactions : avoiding Interaction of elements of different categories with structure conservation : Cyrille Bertelle Emergent Computing 69/80

77 Structure Detection and their management Structures and interactions : breaking Interaction of elements of different categories with structure losing : Cyrille Bertelle Emergent Computing 69/80

78 Dynamical Distribution Computation using Ant System Dynamical agent-based simulation distribution over computer network based on ant system How distribute the previous simulation interacting entities on a computer network: minimizing the communication placing communicating entities on the same computer; dealing with load balancing distributing entities to all the computers, respecting their power capabilities. Proposed solution: an innovative algorithm called AntCO 2, for Ant Competition Colonies. Cyrille Bertelle Emergent Computing 70/80

79 Dynamical Distribution Computation using Ant System Dynamical agent-based simulation distribution over computer network based on ant system How distribute the previous simulation interacting entities on a computer network: minimizing the communication placing communicating entities on the same computer; dealing with load balancing distributing entities to all the computers, respecting their power capabilities. Proposed solution: an innovative algorithm called AntCO 2, for Ant Competition Colonies. Cyrille Bertelle Emergent Computing 70/80

80 Dynamical Distribution Computation using Ant System Dynamical agent-based simulation distribution over computer network based on ant system How distribute the previous simulation interacting entities on a computer network: minimizing the communication placing communicating entities on the same computer; dealing with load balancing distributing entities to all the computers, respecting their power capabilities. Proposed solution: an innovative algorithm called AntCO 2, for Ant Competition Colonies. Cyrille Bertelle Emergent Computing 70/80

81 Dynamical Distribution Computation using Ant System AntCO 2 : principle Extension of the classical ant system with dynamical colored graph, colored pheromons. Each ant is associated to one color: It is attracted by the pheromon of its color. It is repulsed by the pheromons of other colors. Cyrille Bertelle Emergent Computing 71/80

82 Dynamical Distribution Computation using Ant System AntCO 2 : principle Extension of the classical ant system with dynamical colored graph, colored pheromons. Each ant is associated to one color: It is attracted by the pheromon of its color. It is repulsed by the pheromons of other colors. Cyrille Bertelle Emergent Computing 71/80

83 Dynamical Distribution Computation using Ant System AntCO 2 : results (1) Cooperation and Competition process leading to emergent clustering Decentralized process with flexibility and robutness properties Cyrille Bertelle Emergent Computing 72/80

84 Dynamical Distribution Computation using Ant System AntCO 2 : validation r 1 : communication loading (rate of communications between computers over all the communications between entities) r 2 : Computer ressources loading (rate of more busy computer loading over less busy computer loading) Cyrille Bertelle Emergent Computing 73/80

85 Dynamical Distribution Computation using Ant System AntCO 2 : results (2) Cyrille Bertelle Emergent Computing 74/80

86 Economical and Social Modelling Urban car traffic (1) Dynamic urban circulation based on ant system; Urban cartography represented by weighted graph; Edge weighs are dynamic and represented the local traffic load; Ants cross the graph and deposite numerical pheromones to mark optimal solution; Example of ring road first solution which become loaded then a new adaptive solution is dynamically computed. Cyrille Bertelle Emergent Computing 75/80

87 Economical and Social Modelling Urban car traffic (2) Cyrille Bertelle Emergent Computing 76/80

88 Economical and Social Modelling Genetic automata and spatial organizations for Decision Systems Support Automata-based strategy for agent behavior in iterative prisoner dilemma (Cf. Rawan Ghnemat spresentation); Probabilistic automaton for adaptive strategy: The probabilities evolve using a genetic algorithm; The fitness is the accumulated payoff along steps. Perspectives: Integration of ant system and genetic automata over spatial representation to propose an emergent computation leading to a global environment support system. Application: concurrent telecom societies can share collective antenna or compete for exclusive use. Cyrille Bertelle Emergent Computing 77/80

89 Economical and Social Modelling Global Environment Support System Cyrille Bertelle Emergent Computing 78/80

90 Economical and Social Modelling Emotional aspect in Decision making Modelization of the complex interaction emotion-cognition-action Application to learning modelling and to risk management Adaptive behavior or learning processus modelling using genetic automata Emotion modelling using OCC (Ortony, Clore and Colin) model over fuzzy functions (stigmoide) which give transition values of behavior automata. Pluridisciplinary approach: Psychologist scientific researchers are associated and allow the modelling validation with experimental data. Cyrille Bertelle Emergent Computing 79/80

91 Conclusion Conclusion Complex systems conceptual approaches allow to model current world natural and artificial systems, respecting their complexity Computer Science propose nowaday efficient way to modelize these complex systems Toward a generalization of decentralized vision of computing and interacting systems as new challenge in modelling and systems understanding Cyrille Bertelle Emergent Computing 80/80