Blondie25 was a top 12/13 engine back in 2006. Blondie25 was a NN checkers engine (Blondie24) ported over to chess.M ANSARI wrote: ↑Thu Jan 17, 2019 3:07 pmNot sure who is responsible for LC0, but holy crap ... what an incredible achievement! I would have never thought it would be possible to even be in the top 50 engines a few months back! Also what a monster SF has become! It seems like only recently where SF was slowly climbing the engine ladder and now it has become a real beast!
So it was just a matter of time.
"The results of 12 games played on each side were 1 win, 4 losses, and 7 draws for black, and 2 wins, 6 losses, and 4 draws for white. Fritz 8.0 is rated currently at 2752 (plusmn20) on SSDF (the acronym for the Swedish Chess Computer Association), placing it as the 12th strongest program in the world. At the time of the contest between Blondie25 and Fritz 8.0, Fritz 8.0 was rated #5 in the world. The results are the first case of an evolved chess program defeating a world-class chess program (three times). The performance rating for Blondie25 against Fritz 8.0 was 2635.33, which compares well with the previous performance rating of 2650 against Pocket Fritz 2.0. Blondie25 was then tested against a nationally ranked human chess master, rated 2301. In four games, Biondie25 won three and lost one"
"The weights of the neural network were obtained by an evolutionary algorithm (an approach now called neuroevolution). In this case, a population of Blondie24-like programs played each other in checkers, and those were eliminated that performed relatively poorly. Performance was measured by a points system: Each program earned one point for a win, none for a draw, and two points were subtracted for a loss. Points were earned for each neural network after a multiple of games; the neural networks did not know which individual games were won, lost, or drawn. After the poor programs were eliminated, the process was repeated with a new population derived from the winners. In this way, the result was an evolutionary process that selected programs that played better checkers games."
The only difference is that GPUs weren't as strong in the past so it used alpha/beta pruning instead of a MCTS-like algorithm as it had to rely on CPU power. But NN-engines have had top-tier potential for a while now.