卡內基梅隆團隊用 8000 美元的 AI 擊敗了 Stratego 最強玩家
卡內基梅隆、紐約大學、斯坦福大學和麻省理工學院的團隊構建了一個 AI，擊敗了最好的 Stratego 玩家 Pim Niemeijer，訓練成本僅為 8000 美元。這顯示了在不完全資訊遊戲中 AI 的顯著改進。
a team from carnegie mellon university, new york university, stanford university, and mit has developed ataraxos, an ai that has decisively beaten pim niemeijer, widely regarded as the best stratego player of all time, in a 20-game series with 15 wins, one loss, and four draws. niemeijer has won four world championships, 15 dutch national titles, and two online world championships. the ataraxos ai was trained with a budget of less than $8,000, which is significantly less than deepmind's reported $3 million to $4.5 million investment in its deepnash system that failed to surpass top human players. ataraxos learned through self-play without human data, utilizing a custom gpu simulator and high sample efficiency. the ai's success is attributed to a regularization technique that forces it to vary its play and avoid getting stuck in a fixed strategy.