GP-rush

Abstract

We evolve heuristics to guide IDA* search for the 6x6 and 8x8 versions of the Rush Hour puzzle, a PSPACE-Complete problem, for which no efficient solver has yet been reported. No effective heuristic functions are known for this domain, and-before applying any evolutionary thinking-we first devise several novel heuristic measures, which improve (nonevolutionary) search for some instances, but hinder search substantially for many other instances. We then turn to genetic programming (GP) and find that evolution proves immensely efficacious, managing to combine heuristics of such highly variable utility into composites that are nearly always beneficial, and far better than each separate component. GP is thus able to beat both the human player of the game and also the human designers of heuristics.

Genetic Programming Theory and Practice XII

“…Spector and his colleagues earned the competition's top prize twice, once for the use of EC to produce quantum computing results that were published in a top physics journal (Barnum et al 2000;Spector 2004) and once for results in pure mathematics that exceeded human performance by several orders of magnitude (Spector et al 2008). Sipper, who has six wins, tackled a string of hard games and puzzles, evolving game-playing strategies that held their own in competition against humans (Sipper 2006;Hauptman et al 2009;Benbassat et al 2012). In collaborative work with a partner from the semiconductors industry Sipper attained marked improvement over humans in developing automatic defect classifiers for patterned wafers (Glazer and Sipper 2008).…”

“…Over the past 7 years Sipper has done extensive research in the area of games (Sipper et al 2007;Hauptman and Sipper 2005b, a;Hauptman and Sipper 2007b;Azaria and Sipper 2005a, b;Benbassat and Sipper 2010;Hauptman and Sipper 2007a;Hauptman et al 2009;Shichel et al 2005), which culminated in his recent book, "Evolved to Win" (Sipper 2011) (see also www.moshesipper.com/games). Among the games successfully tackled are: chess, backgammon, checkers, Reversi, Robocode (tank-war simulation), Rush Hour, and FreeCell.…”

“…A recent line of research has attempted to build a more general form of evolutionbased game intelligence by employing a structure known as a policy, which is an ordered set of search-guiding rules (Hauptman et al 2009;Elyasaf et al 2012). Policies are complex structures that allow one to define specific conditions under which certain actions are performed.…”

“…For example, Stephanie Forrest's paper (Forrest et al 2009) uses existing human knowledge embedded in non-faulty parts of the code to repair parts of the code that are faulty. Policy based GP (Hauptman et al 2009;Elyasaf et al 2012) is another such area where human knowledge is integrated into an EC system that then evolves a solution that is human competitive. This particular trend certainly suggests a rethink of the artificial (intelligence)-to-(human) intelligence (A/I) ratio, suggested by John Koza et al, which states that GP delivers a high amount of artificial intelligence relative to the relatively minimal amount of human intelligence that is put in to the system (Koza et al 2003).…”

Machine Learning Diffuse Optical Tomography Using Extreme Gradient Boosting and Genetic Programming

“…Therefore, in this article, for the first time, to the best of our knowledge, we use XGBoost to detect tumors located in compressed breast tissues. We later apply an optimization method called genetic programming (GP) [ 34 , 35 , 36 ] to improve our results further. GP is an evolutionary algorithm that uses natural selection to evolve solutions to optimization problems.…”