more info / quasi-blog

This game was inspired by Shannon's "A Mind-reading(?) Machine" and Hagelbarger's "SEER, A SEquence Extrapolating Robot", both from the early 1950's. Both were at Bell Labs at the time. As legend has it, they had their machines compete against each other, and Shannon's won, by a margin of 55-45.
Here is a copy of each paper, in which they describe their machines in detail and even provide schematics.
  • A Mind-reading(?) Machine
  • SEER, A SEquence Extrapolating Robot

    More recent work in online learning involves "A Context-Tree Weighting Method" [Willems et al.] and "How to use expert advice" [Cesa-Bianchi et al.]. The current implementation of the applet uses a variant on the Context-Tree Weighting Method. While this method performs adequately, it can be beaten relatively easily by certain sequences.

    Future work involves combining various kinds of 'experts' using the results in "How to use expert advice". Hopefully this will overcome the limitations of the current algorithm.

    The idea of making the Mindreader into a game belongs to Professors Yoav Freund and Rob Schapire. They also developed the first version of the Mindreader game, upon which the current applet is based.

    View my project presentation and some preliminary data analysis slides here!

    Download the relevant code here. Note the applet is built using the Processing platform.

    Last updated 4.17.06 11:21pm by Anup Doshi
    Back to the game