Chess Engines Do Weird Stuff
This title could be clearer and more informative.Try out Clickbait Shieldfor free (5 uses left this month).
Chess engines like AlphaZero and lc0 reveal surprising ML techniques applicable beyond chess. Distillation from search proves more effective than pure RL, with search contributing ~1200 elo vs model quality's ~200 elo. Runtime adaptation allows networks to adjust evaluations during play. SPSA enables gradient-free optimization
•4m read time• From girl.surgery
Table of contents
Training methodTraining at runtimeTraining on winningTuning through C++Weird architectureNotesSort: