8月25日 的 Poker AI - Game & Strategy 應用程式分析

Texas Hold'em poker is a popular and widely studied problem in the fields of Game Theory and Artificial Intelligence. Algorithms such as Tree Search, Regret Minimization, Reinforcement Learning and Neural Networks have been developed to solve poker. Nevertheless, solving multiplayer no-limit Texas Hold'em in real time remains a major challenge considering its vast combination of possible states and the existence of hidden information. Based on Regret Minimization algorithms, we developed an innovative solving architecture that combines multi-stage Pre-training, Parallel Sampling, Equilibrium Finding, Dynamic Programming, State Abstraction, Encoding-Decoding and Edge Computing. Powered by our built-in AI engine, we present you Poker AI, a fast and accurate Texas Hold'em strategy solver that runs in real time on edge devices with limited computing resources. We are also working on a neural network version of Poker AI using Self-playing, Reinforcement Learning and Deep Neural Networks, which may help avoid errors introduced by abstraction and encoding-decoding, and be more robust in long-tail scenarios that rarely occur. We will release the neural network version when its performance surpasses that of the regret minimization version in our evaluation tests. Key Features of Poker AI: ·Offline and online poker with robots and human players. Talk is cheap, the best way to improve your poker skills is simple and straight: play as much as you can! With Poker AI, you can start a game anytime, anywhere, with or without an internet connection, against bots or humans, and without risking any real money! ·Real-time strategy solver and win rate tools. With our built-in AI engine, you can customize your poker scenarios and calculate strategies based on Nash equlibrium efficiently and intuitively. By combining pre-trained backbone strategies with edge computing optimizations, strategies and win rates for up to 10 players can be solved within seconds.