ai presentation
TRANSCRIPT
ARTIFICIAL INTELLIGENCEARTIFICIAL INTELLIGENCE
11
INTRODUCTIONINTRODUCTION
W.W.W. – a virtual society.W.W.W. – a virtual society. WEB MINING – a challenging activity.WEB MINING – a challenging activity. Web – unstructured data, changes Web – unstructured data, changes
are frequent and rapid.are frequent and rapid.
22
WEB MININGWEB MINING
Process of discovering useful Process of discovering useful information or knowledge from information or knowledge from hyperlink structure, page contents hyperlink structure, page contents and data usage.and data usage.
Three types:Three types:
1.1. Web structure mining.Web structure mining.
2.2. Web content mining.Web content mining.
3.3. Web usage mining. Web usage mining. 33
WEB STRUCTURE MININGWEB STRUCTURE MINING
Discovering knowledge from Discovering knowledge from hyperlinks.hyperlinks.
Pages are ranked according to Pages are ranked according to prestige.prestige.
Pages are social animals and Pages are social animals and hyperlinks are relations.hyperlinks are relations.
44
WEB CONTENT MININGWEB CONTENT MINING
Extracts useful information from the Extracts useful information from the content of web pages.content of web pages.
Pages are classified according to Pages are classified according to their topics, or user’s option.their topics, or user’s option.
Ex. -- <meta> tag used in html Ex. -- <meta> tag used in html scripting.scripting.
55
WEB USAGE MININGWEB USAGE MINING
Aims to automatically discover and Aims to automatically discover and analyze patterns.analyze patterns.
Profiles of users interacting with a Profiles of users interacting with a web site is captured, modeled and web site is captured, modeled and analyzed in order to improve analyzed in order to improve services.services.
Ex. -- private messaging in “orkut” Ex. -- private messaging in “orkut”
66
SWARM SWARM INTELLIGENCEINTELLIGENCE
77
SWARM INTELLIGENCESWARM INTELLIGENCE
Used in representing collective Used in representing collective behavior of decentralized, self behavior of decentralized, self organized artificial systems.organized artificial systems.
Collective behavior can lead to Collective behavior can lead to emergence of apparent intelligent emergence of apparent intelligent behavior.behavior.
SI systems are able to communicate SI systems are able to communicate with each other and to interact with with each other and to interact with their environment.their environment.
88
EXAMPLES OF SI SYSTEMEXAMPLES OF SI SYSTEM
Bird flockingBird flocking
Ant colonyAnt colony
Animal herdingAnimal herding
99
ANT COLONY OPTIMIZERANT COLONY OPTIMIZER
Inspired by ants.Inspired by ants. Ants find path from nest to food by Ants find path from nest to food by
creating a network of pheromone creating a network of pheromone trails.trails.
Similar to ants a www user navigates Similar to ants a www user navigates without the information of route and without the information of route and without having a view of global without having a view of global environment.environment.
1010
BIOLOGICAL INSPIRATIONBIOLOGICAL INSPIRATION
A French entomologist observed that:A French entomologist observed that:
1.1. A significant stimulus is an indirect, A significant stimulus is an indirect, non symbolic form of communication, non symbolic form of communication, mediated by the environment.mediated by the environment.
2.2. Information is local, can be accessed Information is local, can be accessed only by visiting local area in which it only by visiting local area in which it was released. was released.
1111
Swarming – The DefinitionSwarming – The Definition
aggregation of similar animals, aggregation of similar animals, generally cruising in the same directiongenerally cruising in the same direction
Termites swarm to build coloniesTermites swarm to build colonies Birds swarm to find foodBirds swarm to find food Bees swarm to reproduceBees swarm to reproduce
1212
Why do animals swarm?Why do animals swarm?
To forage betterTo forage better To migrateTo migrate As a defense against predatorsAs a defense against predators
Social Insects have survived for millions Social Insects have survived for millions of years.of years.
1313
Swarming is PowerfulSwarming is Powerful
Swarms can achieve things that an Swarms can achieve things that an individual cannotindividual cannot
1414
Collision AvoidanceCollision Avoidance
Rule 1: Avoid Collision with Rule 1: Avoid Collision with neighboring birds neighboring birds
1515
Velocity Matching Velocity Matching
Rule 2: Match the velocity of Rule 2: Match the velocity of neighboring birdsneighboring birds
1616
Flock CenteringFlock Centering
Rule 3: Stay near neighboring birdsRule 3: Stay near neighboring birds
1717
Swarming - CharacteristicsSwarming - Characteristics
Simple rules for each individualSimple rules for each individual
No central controlNo central control Decentralized and hence robustDecentralized and hence robust
EmergentEmergent Performs complex functionsPerforms complex functions
1818
Learn from insectsLearn from insects
Computer Systems are getting Computer Systems are getting complicatedcomplicated
Hard to have a master controlHard to have a master control
Swarm intelligence systems are:Swarm intelligence systems are: RobustRobust Relatively simpleRelatively simple
1919
Applications Applications
Movie effectsMovie effects Lord of the RingsLord of the Rings
Network RoutingNetwork Routing ACO RoutingACO Routing
Swarm RoboticsSwarm Robotics Swarm botsSwarm bots
2020
Particle Swarm Particle Swarm OptimizationOptimization
2121
Particle Swarm OptimizationParticle Swarm Optimization Particle swarm optimization imitates Particle swarm optimization imitates
human or insects social behavior.human or insects social behavior. Individuals interact with one another Individuals interact with one another
while learning from their own while learning from their own experience, and gradually move experience, and gradually move towards the goal.towards the goal.
It is easily implemented and has It is easily implemented and has proven both very effective and quick proven both very effective and quick when applied to a diverse set of when applied to a diverse set of optimization problems. optimization problems.
2222
Bird flocking is one of the best Bird flocking is one of the best example of PSO in nature.example of PSO in nature.
One motive of the development of PSO One motive of the development of PSO was to model human social behavior.was to model human social behavior.
2323
Applications of PSOApplications of PSO
Neural networks like Human tumor Neural networks like Human tumor analysis, Computer numerically controlled analysis, Computer numerically controlled milling optimization;milling optimization;
Ingredient mix optimization;Ingredient mix optimization; Pressure vessel (design a container of Pressure vessel (design a container of
compressed air, with many constraints).compressed air, with many constraints).Basically all the above applications fall in a Basically all the above applications fall in a
category of finding the global maxima of a category of finding the global maxima of a continuous, discrete, or mixed search continuous, discrete, or mixed search space, with multiple local maxima.space, with multiple local maxima.
2424
Algorithm of PSOAlgorithm of PSO
Each particle (or agent) evaluates the Each particle (or agent) evaluates the function to maximize at each point it visits function to maximize at each point it visits in spaces.in spaces.
Each agent remembers the best value of Each agent remembers the best value of the function found so far by it (pbest) and the function found so far by it (pbest) and its co-ordinates.its co-ordinates.
Secondly, each agent know the globally Secondly, each agent know the globally best position that one member of the flock best position that one member of the flock had found, and its value (gbest).had found, and its value (gbest).
2525
Algorithm – Phase 1 (1D)Algorithm – Phase 1 (1D)
Using the co-ordinates of pbest and gbest, Using the co-ordinates of pbest and gbest, each agent calculates its new velocity as:each agent calculates its new velocity as:
vvii = v = vii + c + c11 x rand() x (pbestx x rand() x (pbestxii – presentx – presentxii))
+ c+ c22 x rand() x (gbestx – presentx x rand() x (gbestx – presentx ii))
where 0 < rand() <1where 0 < rand() <1
presentxpresentxii = presentx = presentxii + (v + (vii x x ΔΔt)t)
2626
Algorithm – Phase 2 (n-Algorithm – Phase 2 (n-dimensions)dimensions)
In n-dimensional space :In n-dimensional space :
2727
2828
Ant Colony OptimizationAnt Colony Optimization
2929
Ant Colony Optimization - Ant Colony Optimization - Biological Inspiration Biological Inspiration
Inspired by foraging behavior of ants.Inspired by foraging behavior of ants. Ants find shortest path to food source from Ants find shortest path to food source from
nest.nest. Ants deposit pheromone along traveled Ants deposit pheromone along traveled
path which is used by other ants to follow path which is used by other ants to follow the trail.the trail.
This kind of indirect communication via This kind of indirect communication via the local environment is called stigmergy.the local environment is called stigmergy.
Has adaptability, robustness and Has adaptability, robustness and redundancy.redundancy. 3030
Foraging behavior of AntsForaging behavior of Ants
2 ants start with equal probability of 2 ants start with equal probability of going on either path.going on either path.
3131
Foraging behavior of AntsForaging behavior of Ants
The ant on shorter path has a shorter The ant on shorter path has a shorter to-and-fro time from it’s nest to the to-and-fro time from it’s nest to the food.food.
3232
Foraging behavior of AntsForaging behavior of Ants
The density of pheromone on the shorter The density of pheromone on the shorter path is higher because of 2 passes by the path is higher because of 2 passes by the ant (as compared to 1 by the other).ant (as compared to 1 by the other).
3333
Foraging behavior of AntsForaging behavior of Ants
The next ant takes the shorter route.The next ant takes the shorter route.
3434
Foraging behavior of AntsForaging behavior of Ants
Over many iterations, more ants begin Over many iterations, more ants begin using the path with higher pheromone, using the path with higher pheromone, thereby further reinforcing it.thereby further reinforcing it.
3535
Foraging behavior of AntsForaging behavior of Ants
After some time, the shorter path is After some time, the shorter path is almost exclusively used.almost exclusively used.
3636
Various AlgorithmsVarious Algorithms
First ACO algorithm.First ACO algorithm. Pheromone updated by all ants in the Pheromone updated by all ants in the
iteration.iteration.
Ants select next vertex by a stochastic Ants select next vertex by a stochastic function which depends on both function which depends on both pheromone and problem-specific pheromone and problem-specific heuristic heuristic
3737
Various Algorithms - 2Various Algorithms - 2
Ant Colony System (ACS):Ant Colony System (ACS): Local pheromone update in addition to Local pheromone update in addition to
offline pheromone update.offline pheromone update. By all ants after each construction step only By all ants after each construction step only
to last edge traversed.to last edge traversed. Diversify search by subsequent ants and Diversify search by subsequent ants and
produce different solutions in an iteration.produce different solutions in an iteration. Local update: Local update: Offline updateOffline update: :
3838
Theoretical DetailsTheoretical Details
Convergence to optimal solutions has Convergence to optimal solutions has been proved.been proved.
Can’t predict how quickly optimal Can’t predict how quickly optimal results will be found.results will be found.
Suffer from stagnation and selection Suffer from stagnation and selection bias.bias.
3939
ACO in Network RoutingACO in Network Routing
4040
Ant like agents for routingAnt like agents for routing
Intuitive to think of ants for routing.Intuitive to think of ants for routing. Aim is to get shortest pathAim is to get shortest path Start as usualStart as usual
Release a number of ants from source, let the Release a number of ants from source, let the age of ant increases with increase in hopsage of ant increases with increase in hops
decide on pheromone trails . decide on pheromone trails . Problem – Ants at an node do not know the Problem – Ants at an node do not know the
path to destination, can't cahnge table path to destination, can't cahnge table entryentry
4141
Routing continued ...Routing continued ...
Possible SolutionsPossible Solutions first get to dest. and then retracefirst get to dest. and then retrace
Needs memory to store the pathNeeds memory to store the path And intelligence to revert the pathAnd intelligence to revert the path
Leave unique entries on nodesLeave unique entries on nodes a lot of entries at every nodea lot of entries at every node
Observation – At any intermediate node, Observation – At any intermediate node, ant knows the path to source from that ant knows the path to source from that node.node. now leave influence on routing table having now leave influence on routing table having
entry “route to source via that link”entry “route to source via that link”4242
Routing contd ...Routing contd ...
Now at any node it has information about Now at any node it has information about shortest path to dest., left by ants from shortest path to dest., left by ants from dest.dest.
The ant following shortest path should The ant following shortest path should have maximum influencehave maximum influence
A convenient form of pheromone can be A convenient form of pheromone can be inverse of age + constantinverse of age + constant
The table may get frozen, with one entry The table may get frozen, with one entry almost 1, add some noise f i.e probabilty almost 1, add some noise f i.e probabilty that an ant choses purely random path that an ant choses purely random path
4343
Dealing with congestion Dealing with congestion
Add a function of degree of Add a function of degree of congestion of each node to age of an congestion of each node to age of an antant
Delay an ant at congested node, this Delay an ant at congested node, this prevents ants from influencing route prevents ants from influencing route tabletable
4444
SI - LimitationsSI - Limitations
Theoretical analysis is difficult, due to Theoretical analysis is difficult, due to sequences of probabilistic choicessequences of probabilistic choices
Most of the research are experimentalMost of the research are experimental Though convergence in guaranteed, Though convergence in guaranteed,
time to convergence is uncertaintime to convergence is uncertain
4545
ConclusionConclusion
Provide heuristic to solve difficult Provide heuristic to solve difficult problems problems
Has been applied to wide variety of Has been applied to wide variety of applicationsapplications
Can be used in dynamic applicationsCan be used in dynamic applications
4646