Java Duke with Genetic Algorithm

Here is visual results of applying genetic algorithm for optimization in high-dimensional space. The problem is - to represent some picture, using limited number of semi-transparent polygons. Each vertex of polygon represented by couple of numbers (x, y), and each polygon associated with colour (R, G, B, A). So we can represent each polygon as array of numbers [ (x1, y1), (x2, y2), (x3, y3), (x4, y4), (x5, y5), (x6, y6) (r, g, b, a) ] - 16-dimensional vector. And 120 polygons might be united in one large array (represented as 120*16 = 1920 - dimensional vector). So we can reformulate problem - find vector, which will be transformed into polygons, which approach our picture in the best way (find global minimun in 1920-dimensional non-linear space). Application has to find colour of each polygon, and coordinates of each polygons vertex. The number of possible combinations is extremly large. Just for fun I've tried genetic algorithm to solve this problem, and wrote experimental java application to do that. Average optimizing time is 7-10 min. Generated pictures looks like kind of cubism art :)

Category:

Science & Technology

Tags:

• Genetic Algorithm
• GA
• Genetic Programming
• AI
• Java
• Duke