Presentation is loading. Please wait.

Presentation is loading. Please wait.

EE459 I ntroduction to Artificial I ntelligence Genetic Algorithms Kasin Prakobwaitayakit Department of Electrical Engineering Chiangmai University.

Similar presentations


Presentation on theme: "EE459 I ntroduction to Artificial I ntelligence Genetic Algorithms Kasin Prakobwaitayakit Department of Electrical Engineering Chiangmai University."— Presentation transcript:

1 EE459 I ntroduction to Artificial I ntelligence Genetic Algorithms Kasin Prakobwaitayakit Department of Electrical Engineering Chiangmai University

2 The Role of Search Many (AI) problems can be formulated in terms of searching for a solution in a list of possibilities Chess, Route finding, Bandwidth allocation, Optimize circuit’s topology and values, Determining neural network weights.

3 The Role of Search Usually, each candidate solution has associated with it some measure of how good it is cost function, error measure, RMS error, Other error

4 An Example 123454321 123456789 1 234565432 2 345676543 3 456787654 4 567898765 5 456787654 6 345676543 7 234565432 8 123454321 9 flour sugar Consider the problem of choosing the correct amount of sugar and flour to produce the best quality biscuit This grid shows the axes are amount of flour amount of sugar giving quality of biscuit

5 Exhaustive Search Evaluate each combination in turn using some systematic method, e.g. start(1,1)  (1,9) (2,1)  (2,9) … (9,1)  (9,9) sometimes called the ‘brute-force’ approach This cannot be done for most real problems there are too many combinations too computationally expensive

6 Hill Climbing The hill climbing algorithm start at either a fixed or random position evaluate the current position at each step evaluate in each of the surrounding directions –up, down, left, right move in direction of greatest improvement stop if all moves are lower than current position This is guaranteed to find the ‘peak’ (maximum) but only if there is just one (global) maximum

7 Local Maxima 123212321 123456789 1 232121232 2 321232123 3 212343212 4 123454321 5 212343212 6 321232123 7 232121232 8 123212321 9 flour sugar Suppose the quality grid changes to this If the hill-climbing starts at e.g. (5,1) and heads right first the global maximum at (5,5) is found However, if the hill- climbing starts at e.g. (5,1) and heads up first a local maxima at (3,1) is found

8 Monte Carlo Suppose we just guess random positions The ‘Monte Carlo’ algorithm generate a random stating position evaluate the starting position and store it as best repeat generate a new random position evaluate the new position if the new position is better than the best found so far –store the new position as the best until we decide to stop (e.g. not improved for 20 goes) This may or may not find the global maximum

9 Evolutionary Methods EA’s use nature as an inspiration the principles of evolution birth and death of individuals in a population survival of the fittest multiple generations mating or ‘crossover’ (two parents creating one offspring) mutation (random variation) There are many variations of the EA model genetic algorithms evolution strategies evolutionary / genetic programming

10 Genetic Algorithms Genetic algorithms (GA’s) were originated by John Holland, University of Michegan, 1970’s GA’s use the following principles a population contains multiple individuals each individual comprises a fixed-length string a string of genes is a chromosome crossover and mutation generate new individuals operate on the chromosomes fitness used to choose survivors at each generation

11 Evolution Strategies Evolution strategies (ES’s) were originated by Rechenberg, Schwefel and Bienert at the Technical University of Berlin in the mid 1960’s ES’s use the following principles direct floating point representation of parameters apply mutation by changing these parameters according to normal probability distributions small changes most likely large changes unlikely (but still possible) originally just one parent and one offspring since extended to multiple individuals and maybe crossover


Download ppt "EE459 I ntroduction to Artificial I ntelligence Genetic Algorithms Kasin Prakobwaitayakit Department of Electrical Engineering Chiangmai University."

Similar presentations


Ads by Google