5/18/2009. CS6800 Presentation I Thap Panitanarak. Three branches of natural computing

Transcription

1 CS6800 Presentation I Thap Panitanarak Usefulness of computers in many applications/area Most computers are static do whatever the order is Decades of research Evolve in time In past 2-3 decades, nature + computing Natural science + Computing science Process of extracting ideas from nature to develop computational systems, or using natural materials to perform computation Three main goals Find new problem solving techniques Synthesize natural phenomena Construct novel computing devices using natural materials (in addition to silicon) Three branches of natural computing 1. Computing inspired by nature Nature inspiration for new computational techniques 2. Simulation and emulation of nature by means of computing Synthetic process for creating patterns, forms, behaviors & organisms to mimic various natural phenomena 3. Computing with natural materials Use natural materials, not silicon-based, to perform computation Require knowledge from various fields Physics Chemistry Biology Engineering Computer science Mathematics Etc 1

2 List of some well-known research Computing inspired by nature Artificial neural networks Evolutionary computing Swarm intelligence Artificial immune system Synthesis of natural phenomena Fractal geometry Artificial life Computing with new materials DNA computing Quantum computing Computing inspired by nature The oldest & most popular among three Goals Devise theoretical models & implemented in computers Provide (alternative) techniques/algorithms for solving complex problems Other terms Bio-inspired computing Biologically motivated computing Computing with biological metaphors Artificial neural networks MC Culloch & Pitts First mathematical model of a neuron Artificial Neural Networks ANNs Inspired by nervous system - human brain Problem solving: Input Algorithms (?) output Different from computational neuroscience Biological Motivation Brain Connection between massive number of neurons basic units used for computation (also in ANNs with simplified abstract models) Connection Synapse, a small junction Forward / feedback forming networks Cognitive abilities e.g. perception, thinking and inferring Representation of information/knowledge in a distributed way with parallel processing Concept of learning & experiencing Abilities to modify & update itself Design Principles Neurons, nodes processing elements Receive / send stimuli exchanging information Neuron network connection among neurons Synapses information transmitted between neurons Strength, weight value efficiency of a synapse Learning weight adjusting 2

3 Characterized by Three Features 1. Artificial neurons Weight, summing junction & activating function 2. Architecture / structure of network Layers input, hidden, output Feed forward vs. recurrent 3. Training / learning algorithms Supervised, unsupervised (self-organized) & reinforcement (In principle,) compute any computable function Clustering Classification Pattern recognition Mapping problems Evolutionary Computing Evolutionary biology Search & optimization techniques for solving complex problems Population reproduction genetic variation selection new population Increasingly fitter to their environment Biological Motivation Charles Darwin evolution & natural selection Evolutionary biology diversity of life, differences & similarities, characteristics of organisms Evolving system - one generation after other Current generation Reproduction & Evolution Next generation Natural selection Survivors Next generation - Genetically changed, genes in chromosomes Evolution - Mutation & crossing-over Survivors - Fitted to environment & advantage over others (better) Design Principle Standard evolutionary algorithm Population of individuals that reproduce with inheritance allow to reproduce Genetic variation mutation & crossing-over Natural selection adaptability & fitness values Generic, iterative & probabilistic Maintain the same population size Individual = (encoded) potential solution Genetic algorithms 3

4 Genetic algorithms (pseudo code) 1. Choose initial population (randomly generated) 2. Evaluate the fitness of each individual in the population 3. Repeat until termination: (time limit or sufficient fitness achieved) 1. Select best-ranking individuals to reproduce 2. Breed new generation through crossover and/or mutati on (genetic operations) and give birth to offspring 3. Evaluate the individual fitnesses of the offspring 4. Replace worst ranked part of population with offspring Planning - routing, scheduling and packing Design - signal processing Simulation & identification Control Classification - machine learning, pattern recognition & classification Swarm intelligence is a property of systems of unintelligent agents of limited individual capabilities exhibiting collectively intelligent behavior. Two main lines of research Social insects Human societies Main characteristic Dealing with a population of individuals capable of interacting with the environment & one another Ant colony optimization Experiments of Goss et al. & Deneubourg Argentine ant Iridomyrmex humilis Find the shortest path between food source & nest Indirect communication using pheromone, called stigmergy By laying & following pheromone, the shortest is discovered Traveling salesman problem? Artificial ants & Artificial pheromone trails The shorter tour, the more pheromone Decoy rate of pheromone In some iteration steps, the best tour has most artificial ants Find paths to goal 4

5 Particle swarm optimization Simulation of human social behavior Ability of human societies to process knowledge Individuals search for solution by own experience & experience of others Evaluate, compare & imitate Interaction among individuals (some of neighborhoods) & environment with ability to process knowledge leads to optimal solution Best solution represented in a point or surface in d-dimensional space Individuals as points in space move to solution using their updated velocities Scope & Application discrete optimization problems Example TSP Network routing Graph Coloring etc Artificial Immune Systems (AIS) Adaptive systems inspired by theoretical and experimental immunology with the goal of solving problem Relatively young research Various types of immune algorithms Biological Motivation Immune systems Protect our bodies against infections caused by pathogens (e.g. viruses, bacteria, fungi & parasites) Two major immune systems Innate immune system Determine pathogens Send chemical signal to other immune cells (including those in Adaptive IS) to start fight against pathogens & to stop whence infections have been eliminated Adaptive immune system Determine some pathogens not known by Innate IS Fight & Learn how to fight from infections Three well-known theories Self/Non-Self Discrimination Differentiate between antigens & body itself Initiating appropriate immune response for foreign bodies, whilst leaving everything else untouched Negative selection Destroy danger immature immune cells Clonal Selection & Affinity Maturation Immune cells can be reproduced & mutate Immune cells with high affinity to antigens will be promoted (increased reproductive rate & decreased mutation rate) Immune Network Network of immune cells & molecules Interact with each other & environment to stimulate recognizing cells, even no antigen invasion Design Principles Vast, but some core idea based on 3 theories Three basic elements based on immune engineering by de Castro & Timmis Representation Elements (immune cells & molecules) Evaluating interactions Among elements & environment Procedures of adaptation Dynamic behaviors 5

6 Machine-learning & pattern recognition Anomaly detection & security of information systems Data analysis (knowledge discovery in databases, clustering, etc.) Agent-based systems Scheduling Autonomous navigation & control Search & optimization Artificial life New tools for synthesize & study of natural phenomena in computers Test biological theories that cannot be tested via traditional experiments & analytic techniques Most cases aim for simulation & emulation Theoretical study Synthetic process Fractal Geometry A rough or fragmented geometric shape that can be split into parts, each of which is (at least approximately) a reduced-size copy of the whole Self-similarity Some natural patterns & structures exhibited in nature Example - ferns, coastlines, mountains, cauliflowers, broccoli, our lungs, our circulatory system, our brains, & our kidneys Modeling & Synthesis Importance issues Fractal dimension Modeling techniques cellular automata L-systems Fractal Dimension Describes fractal complexity of an object How many copies of itself in d-dimensional shape? Reduce factor m = 1/l Cellular Automata Dynamical system that is discrete in both space & time d-dimensional cellular automaton represented by d- dimensional grid Cellular Automaton, C = <S, s 0, G, d, f> S a finite set of state s 0 initial state G cellular neighborhood d dimension of C, d in Z + local transition rule f : S n S, transition function n neighborhood size Neighborhood, G i = {i, i+r 1, i+r 2,..., i+r n } Conf. at time t, C(t) = (s 0 (t), s 1 (t),..., s N 1 (t)) 6

7 Example: one-dimensional cellular automata Lindenmayer systems (L-systems) A. Lindenmayer Development at multicellular level division, enlargement, differentiation & death of cells Modeling of plants L-system grammar, G = {V, S, ω, P} V a finite set of (variable) symbols, alphabet S a finite set of (fixed) symbols, constant ω nonempty word (axiom), ω in V + P a finite set of productions, P subset of V x V * Example: Sierpinski triangle Variables : A B Constants : + Start : A Rules : (A B A B) (B A+B+A) Angle : 60 Description A & B mean both "draw forward", + means "turn left by angle means "turn right by angle" Example: Fractal plant Variables : F X Constants : + Start : X Rules : (X F-[[X]+X]+F[+FX]-X) (F FF) Angle : 25 Description F means "draw forward - means "turn left by angle + means "turn right by angle" Models of natural & non-natural patterns Plant growth, crystal growth, formal language theory, music composition, etc Models of dynamical behavior of many real complex systems Physical fluid, neural networks, traffic flow, etc 7

8 Artificial Life Study of man-made systems that exhibit behaviors characteristic of natural living systems Complement traditional biological sciences concerned with the analysis of living organisms by attempting to synthesize life-like behaviors within computers & other artificial media Contribute to theoretical biology by locating lifeas-we-know-it within the larger picture of lifeas-it-could-be Some Alife Projects StarLogo programming language Behavior simulation in ants, termites & traffic jams Different aspect in swarm intelligence Interactions among turtles & patches in active world Example: Ant simulation Boids Motion simulation in flocks, herds & schools Three rules Collision avoidance & Separation Velocity matching & Alignment Flock centering or Cohesion Example: Bird simulation (Mainly) simulating biological phenomena Generate & control complex behaviors Entertainment & scientific purposes Evolution of Computing gears relays valves transistors integrated circuits silicon ship Other types? New generation of computing DNA & Quantum L. Adleman in 1994 DNA had computational potential DNA can store massive series of text, {A,C,T,G} Solving TSP with 7 cities Biological Motivation DNA - deoxyribonucleic acid contain genetic instructions/information Segments = genes 2 long polymers of simple units called nucleotides Nucleotide - strand Backbone = sugars & phosphate groups Sequence of four bases along backbone Adenine, Guanine, Cytosine & Thymine (A,G,C,T) Binding: A with T, G with C 8

9 Biological operations (also used in DNA computing) Denaturation: separates DNA strands Annealing: fuse DNA strands Polymerase extension: fill in incomplete strands Nuclease degradation: shorten DNA molecules Endonucleases: cut DNA molecules Ligation: link DNA molecules Modifying nucleotides: insert or delete short subsequences Amplification: multiply DNA molecules - polymerase chain reaction (PCR) Gel electrophoresis: measure the length of DNA molecules & separates them by length Filtering: separate or extract specific molecules Synthesis: create DNA molecules Sequencing: read out the sequence of a DNA molecule Filtering Model Filtering out undesirable results Adleman s test-tube computer Algorithm Step 1: generate random paths through the graph Step 2: keep only those paths beginning with ν in & ending with ν out Step 3: if the graph has n vertices, then keep only those paths that enter exactly n vertices. Step 4: keep only those paths that enter all the vertices of the graph at least once. Step 5: if any path remains, say YES; else, say NO Translate the algorithm to molecular biology Vertex single strand sequence of nucleotides of length 20, v i = a i b i, v j = a j b j Edge complement of sequence b i a j Use PCR to generate (huge) desirable amounts of vertices & edges Put them in test-tube, with some chemical agents Massively parallel chemical reactions lead to existence of a solution (if exists) All steps involved with biological operations Result desired strands of length 140, starting with v in, ending with v out & consisting of sequence of all 7 cities Solving NP-complete problems Cryptographic problems Matrix addition & multiplication Parallel machines Quantum physics Theory that explains behavior of objects with atomic scales Quantum mechanics Mathematical framework that describes quantum physics More detail about quantum computing Keith Kelly 9

10 Conclusion Think naturally about computation Think computationally about nature Novel approaches Evolution in computing Inter- & multidisciplinary research References L. N. de Castro, fundamentals of natural computing: an overview, Physics of Life Reviews, 4 (2007), Wikipedia, Question What are the three branches of natural computing? Give two fields of each. Answer Computing inspired by nature Artificial Neural Networks Swarm Intelligence Synthesis of natural phenomena Artificial Life Fractal Geometry Computing with new materials DNA Computing Quantum Computing 10