Prof. Habiba Drias Laboratoire de Recherche en Intelligence Artificielle LRIA Computer Science Department USTHB Algiers Algeria

Transcription

1 Swarm Intelligence and Evolutionary Algorithms Habiba Drias Wikipedia.org Prof. Habiba Drias Laboratoire de Recherche en Intelligence Artificielle LRIA Computer Science Department USTHB Algiers Algeria 1

2 An unprecedented necessity of Large Scale Computing For Sciences and Technologies Quantum Chemistry Bio-Technologies Climatology Nanoscience and Nanotechnologies Bioinformatics Astrophysics Environmental Sciences Nanotechnologies nearly used for repairing the human brain? 3

3 Huge and Complex Data Prohibitive volumes of information Sciences and Technologies have produced huge volumes of data these last decades from all domains: medicine, agriculture, business, education... Artificial intelligence has contributed in accelerating the rhythm of discoveries in these areas. High Performance Hardware has contributed for the treatment and storage of data. Advanced Computer languages helped the software design. 4 Consequences Positive effects Immense progress in all domains. Availability and rapid access to information. Multiple exchanges and interaction. Side effects failures in large-scale distributed environments. A significant increase in complexity of predefined applications.

4 Large Scale Computing An essential issue Traditional Complex Problems Many problems cannot be solved with traditional Hardware and Software Web Emergence Design of a plethora of Web technologies in the last decades Little applications remain local and do not need Web Technologies Consequences New Challenges for computer science Solutions New Hardware Design: Supercomputers, Grids, Clouds, New Software Design: New Computing Paradigms 5

5 New Computing Paradigms Swarm Intelligence Quantum Computing DNA Computing 6

6 Agenda Computational Complexity Issues NP-Completeness NP-Hardness Examples Swarm Intelligence and Evolutionary Algorithms Applications Domains What is Swarm Intelligence? Self-organization Stygmergy Traditional Sources of Inspiration Genetic Algorithm Ant Colony Optimization Particle Swarm Optimization New Sources of Inspirations Bat Algorithm Harmony Search Firefly Algorithm From Natural Phenomena to Swarm and Evolutionary Algorithms Solution representation, fitness function Search Space : Continuous, discrete A Case Study: Bee Swarm Optimization (BSO) 7

7 Computational Complexity Issues NP Class It contains all problems solved with a Nondeterminist Polynomial algorithm P Class It includes all problems solved by algorithms with polynomial time complexity. NP-Completeness It concerns the decision problems with Yes/No answer. NP- complete problems belong to the NP Class. an NP-complete problem has yet no algorithm with a polynomial time complexity. NP-Completeness is an indicator of the algorithm that cannot be solved with realistic resources. NP-Hardness an NP- Hard problem is at least as hard as an NP- Complete problem. 8 P NP-Complete Unless P = NP NP

8 Computational Complexity Issues Problems belonging to the P class are solved efficiently There exists an algorithm of polynomial time complexity for each of them. NP-complete Problems are not solved in an efficient way All algorithms that solve them have at least an exponential time complexity unless P=NP. Incomplete algorithms yield approximate or near-optimal solutions. How exponential growth surpasses both linear and cubic growth. Search time Scatter search Genetics algorithms Problem size 9

9 NP-hard Problems : Examples The Traveling Salesman Problem Collecting schoolchildren Job Scheduling Manufacture engineering The emergency vehicles management Hospital ambulances Timetabling Problem For examination For courses For excursions Satisfiability Problem Satisfiability of logic formulas Reasoning Exponential time complexity or Exponential space complexity or both Natural Language Processing Comprehension Translation Current machines can not handle this category of problems, especially when the data size is very large. 10

10 Traveling Salesman Problem (TSP) Combinatorial Optimization Problem Definition1: Decision problem Instance : a set of cities and the distances that separate each pair of them. Question : Is there a route starting from one city, that includes all the other cities only once and returns to the initial city, such that the whole distance is less than a positive number k? Answer: Yes/No NP-Complete problem Definition2: Optimization problem Instance : same as for Definition1. Question : Find a route starting from one city that includes all the other cities only once and returns to the initial city with a minimum distance. Answer: a route NP-Hard problem 11 MathWorks blogs

11 TSP is NP-Hard Traditional TSP techniques are not capable to yield results in a reasonable time when the number of cities is very large. One possible solution Tackle the problem using artificial intelligence tools such as swarm intelligence. 12

12 Swarm Intelligence: Applications Combinatorial Optimization Traveling Salesman The ability of ants to find the shortest path has created a powerful system: ACO. Scheduling and Planning Modern Transportation Telecommunications Antenna Placement Problem (APP) The Location Management Problem (LMP): Tracking the mobile users activity across a network (nowadays hot topic) Numeric Functions Optimization 13

13 Swarm Intelligence: Applications Artificial Intelligence Applications Natural Language Processing Image Processing Robotics Roboticists studying the cooperative transportation in ants to design effective technical control of a group of robots. Entertainment Industry Movies Games Industry Task Allocation A painting booths programming technique is inspired by the flexible allocation of tasks among bees. 14

14 Swarm Intelligence: Applications Data Mining A new banking data analysis approach is copied on how ants sort their larvae. Routing in Networks A new method of modifying the traffic of a saturated communications network is modeled on the behavior of ants in search of food (Ant Net System). Web Applications Web search and Filtering Web Security 15

15 Swarm Intelligence A promising field of artificial intelligence. Its basic principle and philosophy: Observation from Nature of intelligent phenomena and especially group behaviors. Simulation of collective behaviors of social insects as ants and bees but also animals, fishes and birds: Collective Intelligence. Researchers and more precisely biologic psychologists, undertook deeper investigations on the social interaction of autonomous agents when working together to achieve a certain goal. 16

16 Swarm Intelligence Biological species, from bacteria to humans use social interaction for their surviving. In the group, the information communication influences the individual behaviors and guide all the agents towards the contribution of achieving the same global goal representing the swarm goal. A new paradigm was born under the name of Swarm Intelligence (SI) dedicated to Cooperative Problem Solving (CPS). A rich repertoire of Powerful methods to solve complex problems has been developed. E. Bonabeau, M. Dorigo and G. Theraulaz Swarm Intelligence: from Natural to Artificial Systems, Oxford University Press,

17 What is Swarm Intelligence? Swarm Intelligence The insect or animal societies become models for computing. The collective intelligence is the key idea. Interesting Characteristics Decentralized control Each individual has a local view of its environment but nothing about the whole system. Self organization Stigmergy Robustness and flexibility 18

18 Examples of Natural Groupings A group consists of a number of autonomous agents, coordinating with each other for their survival in a natural environment. Bird Flocking Habiba Drias Wikipedia.org Ant Colony Animal Herd Fish Schooling 19

19 Bird Flocking from Milan Habiba Drias Milan, 12 December

20 Group Characteristics A Group of simple Agents An agent lacks any memory, intelligence or awareness of each other. It performs very simple tasks. A Distributed System No central Control A Complex Global Behavior Global goals: Predator avoidance, Feeding, building a nest, Different tasks corresponding to different goals. The global behavior is hard to model. A Robust System no failure even when some agents do not accomplish their task. A Flexible System The agents adapt to the environment, even in case of disruption. 21

21 Perception Agent Architecture Agent Stimulus + Internal State Behavior Action Environment 22

22 Various ways of Inspiration Bacterial growth Ant colonies Bird flocking Fish schooling Animal herding Sarotiko LinkedIn 23

23 Self Organization Exists in a Decentralized group of autonomous agents With no supervision. An agent performs a simple task, which can be simulated by an automaton. The distribution is over all the agents. Google.com It is a Foundation existing in the group described as: An order emerging from an initial chaos after interactions between agents. It is a Complex Phenomenon Difficult to grasp. hard to model. Emergence Simple agents + Interactions Complex Intelligent System + Order 24

24 Self Organization The Self Organization is based on four principles: Positive feedback made possible by the recruitment of agents to search for a food source to build a nest Negative feedback due to poor environment in food sources Harmful pollution Amplification of fluctuations due to random events Random walk Multiple interactions of the agents, which can be direct or indirect. Pheromone (ants) Dance (bees) 25

25 Stigmergy introduced by Pierre-Paul Grassé in 1959 to refer to termite behavior. A colony of ants traced by pheromone Wikipedia.org Indirect Coordination between simple agents An agent performed an action in the environment to stimulate the performance of the next action. the ants left pheromone on the ground to indicate to the other ants the direction of the food source Different types of bee Dance the bees performed a vigorous dance to guide their congeners towards the region of rich nectar Stigmergy Stuff Self-Organization 26

26 Ant Sygmergy Experience of GAUSS 1989 Food source Food source Ants Nest Ants Nest The ants start exploring the region where they live The majority of ants take the shortest way Ant Stygmergy = Pheromone Dorigo, M. and Stützle, T. (2004), Ant Colony Optimization, MITPress. 27

27 Particle Motion Each particle movement is influenced by: The particle local best position depending on: The particle speed The best performance of the particle neighbors The best known positions of the group. The overall best performance Three types of particles behaviors may exist: Selfish: The particle goes its own way. Conservative: the particle conserves its position. Panurgian: The particle follows the best of all the group. 28

28 Traditional Sources of Inspirations Biology Genetic algorithms -GA- Holland, J.H.: Adaptation in Natural and Artificial Systems. University of Michigan Press, Ann Harbor (1975) Memetic Algorithms -MA- Moscato, P. : On Evolution, Search, Optimization, Genetic Algorithms and Martial Arts: Towards Memetic Algorithms, Caltech Concurrent Computation Program (report 826), (1989). Mathematics Scatter Search -SS- Laguna, M. and R. Martí :Scatter Search Methodology and Implementations in C, Kluwer Academic Publishers, Boston, (2003). Birth of the Evolutionary Algorithms and meta-heuristics 29

29 Traditional Sources of Inspirations For Swarm Algorithms Biology and Nature Ant Colony Optimization -ACO- Dorigo, M.: Optimisation, Learning and Natural Algorithms, PhD thesis, Politecnico di Milano, Italy (1992) Bee Swarm Optimization -BSO- H. Drias, S. Sadeg, S. Yahi: Cooperative Bees Swarm for Solving Max-W-SAT, LNCS Springer, , (2005) Particle Swarm Optimization -PSO- Kennedy, J. and Eberhart, R. Particle swarm optimization, IEEE Int. Conf. on Neural Networks, NJ, , (1995) Essentially from Nature and Biology Nature and Bio-Inspired Approaches 30

30 Other Traditional Sources of Inspirations Bacteria Foraging Optimization -BFO- Passino, K.M.: Biomimicry of Bacterial Foraging for Distributed Optimization and Control. IEEE Control Systems Magazine, (2002) Artificial Immune System -AIS- De Castro, Leandro N.; Timmis, Jonathan, Artificial Immune Systems: A New Computational Intelligence Approch, Springer, 57 58, (2002) Kephart, J. O., A biologically inspired immune system for computers, Proceedings of Artificial Life IV, MIT Press, , (1994) 31

31 New inspiration sources Nature Bat Algorithm -BA- Yang, X. S. A new metaheuristic bat-inspired algorithm, NICSO 2010, (Eds. J. R. Gonzalez et al.), Springer, SCI Vol. 284, 65-74, (2010) Firefly -FS- Xin-She Yang and Xingshi He, (2013). Firefly Algorithm: Recent Advances and Applications, Int. J. Swarm Intelligence, Vol. 1, No. 1, pp DOI: /IJSI Music Composition Harmony search -HS- Xin-She Yang, Harmony Search as a Metaheuristic Algorithm, in: Music-Inspired Harmony Search Algorithm: Theory and Applications (Editor Z. W. Geem), Studies in Computational Intelligence, Springer Berlin, vol. 191, pp (2009) Some sources are out of the scope of Nature and Biology 32

32 From Natural Phenomena to Evolutionary Algorithms Agent Living World Natural Phenomenon A natural agent : ant, bee, bird, fish, The natural environment: forest, prairie, steppe, Evolutionary Algorithm An artificial agent: a potential solution for the problem The search space : The whole set of potential solutions (positive or negative) Group A group of natural agents A population or a set of potential solutions Target Natural Objectives: Food An optimal solution behavior Agent Performance Agent tasks: Searching for food Enemy Avoidance Building nest Sorting larvae Adaptation in the environment 33 A programming technique simulating the natural agent task Fitness function value of the solution

33 Search Space and Evolutionary Algorithms An evolutionary algorithm is an algorithm that considers a population of solutions from the search space and evolves them towards optimal ones. Usually, it simulates the evolution of nature and bio-inspired species and other phenomena. Depending on the type of problem data, the Search Space can be Continuous For continuous problems Examples of meta-heuristics: PSO BA Discrete For Combinatorial problems Examples of meta-heuristics: Genetic Algorithms ACO The majority of evolutionary algorithms were adapted from Continuous to discrete search space, and vice versa. 34

34 Metaheuristics and Swarm intelligence A heuristic is an information or a knowledge on the problem to solve, used in an algorithm to help building a solution more efficiently. It is usually not easy to design, as it depends on the problem. A good mastering of the problem is necessary. A metaheuristic is a generic approach that uses heuristics to solve complex problems. It follows an algorithmic framework dictated by a specific technique. Local Search: evolution of one solution at a time to reach the optimum. Tabu search: a local search with subtle strategies (intensification and diversification). A nature and bio-inspired approach is considered also as a metaheuristic. It follows an evolutionary algorithm schema, dictated by the simulation of the natural species evolution. A metaheuristic refers to a master strategy that guides and modifies other heuristics to produce solutions beyond those that are normally generated in a quest for local optimality, Fred Glover and Manuel Laguna 35

35 Two categories of Metaheuristics Swarm algorithms can be seen as meta-heuristics based population 36

36 A tradeoff between exploration and exploitation All metaheuristics include two search strategies: A Search Intensification: local search in one region of the search space A Search Diversification: exploration of other regions The weight of each search strategy is variable from one meta-heuristic to another. The search types are dictated by the source inspiration, for example: Genetic algorithms: Intensification occurs when the crossover rate is high Diversification is controlled by a high mutation rate PSO: Intensification occurs when the priority is given to the local movement Diversification occurs when the priority is given to global movement Generally: More intensification yields an effective solution. More diversification yields an efficient solution. A balance between intensification and diversification yields an effective and efficient solution. 37

37 Conclusions on Swarm Intelligence Strength Unlike complete algorithms, swarm intelligence algorithms yield a response even if it is not optimal for problems instances of big sizes. More efficient than exact algorithms for complex problems with a large size. Towards Large Scale Computing because of the tremendous volumes of data and complexity of problems: Swarm intelligence may ensure scalability in some degree : the intelligent used scheme mimicking natural species. the random walk performed by the artificial agents. The same architecture can be applied to a couple of agents or thousands of them. The algorithmic framework is flexible, artificial agents can be easily added. The code is reusable for another applications: just the solution representation, the search space are to be thought and written. The system is robust, it continues working in case of some agents failure. 38

38 Conclusions On Swarm Intelligence Limitations Lack in effectiveness The optimal solution is not guaranteed. Solution Assessment. How far is the achieved solution from the optimal one? Parameter Tuning It takes a long time to determine manually the parameters. It is interesting to develop approaches to yield them automatically. Stagnation or premature convergence Some Swarm intelligence algorithms converge in a premature way, yielding a solution with low quality. Palliate to the situation by introducing strategies of intensification and diversification and by determining a right balance between them 39

39 References Bonabeau E., Dorigo M. and Theraulaz G.: Swarm Intelligence: from Natural to Artificial Systems, Oxford University Press, (1999). Cui Z. H. and Cai X. J.: Integral particle swarm optimization with dispersed accelerator information, Fundam. Inform., Vol. 95, , (2009). De Castro L. N. and Timmis J.: Artificial Immune Systems: A New Computational Intelligence Approach, Springer, 57 58, (2002). Davis, M., Putnam, H.: A computing procedure for quantification theory. In: CACM, vol. 7, (1960) Davis, M., Logemann, G., Loveland, D.: A Machine Program for Theorem Proving. Communications of the ACM 5(7), (1962). Dorigo M.: Optimisation, Learning and Natural Algorithms, PhD thesis, Politecnico di Milano, Italy (1992). Drias H. & Ibri S., Parallel ACS for weighted MAX-SAT, In proc of Iwann 2003, LNCS 2686, Springer Verlag, Iles Baleares, Spain, July 2003, Drias H., Sadeg S. and Yahi S.: Cooperative Bees Swarm for Solving the Maximum Weighted Satisfiability Problem, LNCS, Springer Verlag: , (2005). Farmer J.D., Packard N. and Perelson A.: The immune system, adaptation and machine learning, Physica D, 2, , (1986). Fogel L.J., Owens A.J. and Walsh, M.J.: Artificial Intelligence through Simulated Evolution, John Wiley, (1966). 40

40 Garey, M. R. and Johnson, D. S.: Computers and Intractability: A Guide to the Theory of NP- Completeness, W. H. Freeman and Co., (1979). Geem Z. W., Kim J. K. and Loganathan G. V.: A new heuristic optimisation: Harmony search, Simulation, Vol. 76(2), 60-68, (2001). Glover F. and Laguna M.: Tabu Search, Kluwer Academic Publishers, Boston, (1997). References Haddad, O. B., Afshar, A. and Mario M. A.: Honey-bees mating optimisation (HBMO) algorithm: a new heuristic approach for water resources optimisation, Water Resources Management, Vol. 20, , (2006). Holland J.H.: Adaptation in Natural and Artificial Systems, University of Michigan Press, Ann Harbor (1975). Johnson D. S.: Approximation algorithms for combinatorial problems, J. Comput. System Sci., 9 (1974), pp Karaboga D. and Akay B.: A survey: algorithms simulating bee swarm intelligence, Artif Intell Rev, 31:61 85, (2009). Kennedy J. and Eberhart R.: Particle swarm optimization, IEEE Int. Conf. on Neural Networks, NJ, , (1995). 41

41 References Kephart J. O.: A biologically inspired immune system for computers, Proceedings of Artificial Life IV: The Fourth International Workshop on the Synthesis and Simulation of Living Systems. MIT Press, , (1994). Kirkpatrick S., Gellat C. D., and Vecchi M. P.: Optimisation by simulated annealing, Science, 220, , (1983). Laguna M. and Martí R.: Scatter Search Methodology and Implementations in C, Kluwer Academic Publishers, Boston, (2003). Mucherino A. and Seref O.: Modelling and solving real-life global optimisation problems with metaheuristic methods, in: Advances in Modeling Agricultural Systems, Springer Optimisation and Its Applications Series Vol. 25, , (2008). Passino K.M.: Biomimicry of Bacterial Foraging for Distributed Optimization and Control, IEEE Control Systems Magazine, (2002). Parpinelli R. S. and Lopes H. S.: New inspirations in swarm intelligence: a survey, Int. J. Bio-Inspired Computation, Vol. 3, No. 1, 1 16, (2011). 42

42 References Wolpert D. H. and Macready W. G.: No free lunch theorems for optimisation, IEEE Transaction on Evolutionary Computation, Vol. 1, 67-82, (1997). Yang X. S.: Nature-Inspired Metaheuristic Algorithms, Luniver Press, (2008). Yang X. S.: Firefly algorithms for multimodal optimisation, Proc. 5th Symposium on Stochastic Algorithms, Foundations and Applications, LNCS, Vol. 5792, , (2009). Yang X. S.: A new metaheuristic bat-inspired algorithm, in: Nature Inspired Cooperative Strategies for Optimization (NICSO), Springer, SCI Vol. 284, 65-74, (2010). Yang X. S., Cui Z. H., Xiao R. B., Gandomi A. H. and Karamanoglu M.: Swarm Intelligence and Bio-Inpsired Computation: Theory and Applications, Elsevier, London, (2013). 43

43 44