TE
科技回声
首页24小时热榜最新最佳问答展示工作
GitHubTwitter
首页

科技回声

基于 Next.js 构建的科技新闻平台,提供全球科技新闻和讨论内容。

GitHubTwitter

首页

首页最新最佳问答展示工作

资源链接

HackerNews API原版 HackerNewsNext.js

© 2025 科技回声. 版权所有。

“Rock Paper Scissors” Trained in Browser AI/ML

38 点作者 GantMan将近 6 年前

2 条评论

jedberg将近 6 年前
FYI, the author trained a model to recognize what symbol your hand is making, not a model for ideal play.<p>Identifying the hands is pretty slick, but I&#x27;d actually find a model of ideal play to be even slicker, because there is no &quot;optimal strategy&quot; for RPS, but there is an &quot;optimal strategy&quot; against each individual opponent. Making an AI that can learn that strategy would be pretty impressive.
评论 #20627852 未加载
评论 #20629142 未加载
评论 #20627578 未加载
评论 #20627688 未加载
评论 #20632377 未加载
评论 #20628521 未加载
评论 #20628821 未加载
评论 #20628575 未加载
graycat将近 6 年前
Can approach RPS (Rock, Paper, Scissors) strategy as a relatively simple case of classic von Neumann-Morgenstern two person game theory. That theory follows from the classic duality result in linear programming.<p>The main result for players Red and Blue for RPS is: Red plays each of Rock, Paper, Scissors with probability 1&#x2F;3rd and independent of everything else in the known universe. Can do this by using, say, an ordinary die with six sides. Then just from the strong law of large numbers, in the long run Red and Blue will both break even, no matter what Blue did, does, or will do.<p>To win, Red needs some means of predicting Blue&#x27;s play -- no good predictions, no winning. I.e., for Red to win, Blue has to be predictable in some sense.