arxivcs.LGcs.AIcs.GT2026-07-07
A Gold-Standard Study of What Makes a Lightweight Game-Playing Agent Strong
Nima Kelidari, Mohammadsaeed Haghi, Mahdi Salmani
Reinforcement learning agents for imperfect-information card games are only as strong as the opponents they train against, and they are hard to grade, since they beat a random opponent over 99 percent of the time and only tie copies of themselves. So we build a strong, fixed, rul…