Re:DFS, poker, basketball, Starcraft and strong opinions, we...
by gwern on Wednesday December 25, 2019 @05:45PM
(#21880912)
Another one you might like is the Amazonism 'disagree and commit': https://www.amazon.com/p/feature/z6o9g6sysxur57t
I ( https://www.gwern.net/Timing#try-try-again-but-less-less ) see it as a kind of bandit as well, but specifically, Thompson sampling, because you can interpret groups of people with strong but inconsistent beliefs individually as collectively implementing a Bayesian distribution, and then individuals going off and following what seems to then like the most profitable opportunity is equivalent to Thompson sampling: https://people.csail.mit.edu/pkrafft/papers/krafft-thesis-fi...
Re:DFS, poker, basketball, Starcraft and strong opinions, we...
by gautamcgoel on Wednesday December 25, 2019 @03:03PM
(#21880157)
I like these notes on multi-armed bandits (https://arxiv.org/pdf/1904.07272.pdf), I think they're pretty accessible. Section 1.3.3 discuss the OFU principle. Happy holidays to you as well!
I see this Coursera course on OFU: https://www.coursera.org/lecture/practical-rl/optimism-in-fa... and what looks like Auer's original UCRL paper http://papers.nips.cc/paper/3052-logarithmic-online-regret-b... Do you have a preferred source for learning about this? Thanks and Merry Christmas!