5 ms·

Bandit Algorithms

• ivan_ah 8y ago
  Error establishing a database connection ^ It seems we're in the broken-DB arm of the experiment! Archive link: https://web.archive.org/web/20180728075957/http://banditalgs.com/ https://web.archive.org/web/20180728075957/http://banditalgs... Book link: http://downloads.tor-lattimore.com/banditbook/book.pdf http://downloads.tor-lattimore.com/banditbook/book.pdf
• slaymaker1907 8y ago
  Seems to be having some issue but you can view the article here https://web.archive.org/web/20180729114259/http://banditalgs.com/ https://web.archive.org/web/20180729114259/http://banditalgs...
• pronoiac 8y ago
  I hadn't heard of this before! Bandit problems, from Wikipedia[1]: "[imagine] a gambler at a row of slot machines (sometimes known as "one-armed bandits"), who has to decide which machines to play, how many times to play each machine and in which order to play them, and whether to continue with the current machine or try a different machine." [1] https://en.wikipedia.org/wiki/Multi-armed_bandit https://en.wikipedia.org/wiki/Multi-armed_bandit
  • AstralStorm 8y ago
    This is part of statistics 201 and required reading for anyone working on AI as it is closely related to Markov processes, which are used to define Hidden Markov Models. Likewise, Restricted Boltzman Machines which are often used as input layers to modern deep neural and deep belief networks.
• orasis 8y ago
  Just use Bayesian bandits for most binary success/fail problems. They’re very robust to a lot of weird conditions.