Machine Evolution Evolutions Generations of descendants Production of

Machine Evolution

Evolutions § Generations of descendants – Production of descendants changed from their parents – Selective survival § Search processes § Searching for high peaks in the hyperspace 2

Applications § Function optimization – The maximum of a function – John Holland § Solving specific problems – Control reactive agents – Classifier systems § Genetic programming 3

A program expressed as a tree 4

A robot to follow the wall around forever § Primitive functions : AND, OR, NOT, IF § Boolean functions – – AND(x, y) = 0 if x = 0; else y OR(x, y) = 1 if x = 1; else y NOT(x) = 0 if x = 1; else 1 IF(x, y, Z) = y if x = 1; else z § Actions – North, east, south, west 5

A robot to follow the wall around forever § All of the action functions have their indicated effects unless the robot attempts to move into the wall § Sensory inputs : : : n, ne, e, s , sw, w, nw § 만약 함수의 수행결과가 값이 없으면 중 지 6

A robot in a Grid World 7

A wall following program 8

The GP process § Generation 0 (0세대): start with a population of random programs with functions, constants, and sensory inputs – 5000 random programs § Final : Generation 62 60 steps 동안 벽에 있 는 방을 방문한 횟수로 평가 32 cells이면 perfects; 10곳에서 출발하여 fitness 측정 9

Generation of populations I § (i+1)th generation – 10%는 i-the generation에서 copy 5000 populations에서 무작위로 7개를 선택하여 가 장 우수한 것을 선택 (tournament selection) – 90%는 앞의 방법으로 두 프로그램(a mother, a father)을 선택하여, 무작위로 선정한 father 의 subtree를 mother의 subtree에 넣는다 (crossover) 10

Crossover 11

Evolving a wall-following robot § 개별 프로그램의 예 – (AND (sw) (ne)) (with fitness 0) – (OR (e) (west) (with fitness 5(? )) – the best one : : : fitness = 92 (어떤 때) 13

The most fit individual in generation 0 14

The most fit individuals in generation 2 15

The most fit individuals in generation 6 16

The most fit individuals in generation 10 17

Fitness as a function of generation number 18