formal learning theory in context - princeton.edu.pdfVIP

Formal Learning Theory in Context∗ Daniel Osherson Scott Weinstein Princeton University University of Pennsylvania June 15, 2007 Introduction One version of the problem of induction is how to justify hypotheses in the face of data. Why advance hypothesis A rather than B — or in a probabilistic context, why attach greater probability to A than B? If the data arrive as a stream of observations (distributed through time) then the problem is to justify the associated stream of hypotheses. Several perspectives on this problem have been developed including Bayesianism (Howson and Urbach, 1993) and belief-updating (Hansson, 1999). These are broad families of approaches; the citations are meant just as portals. Another approach is to attempt to justify the present choice of hypothesis by situating it in a strategy with good long-term prospects. Such is the idea behind Formal Learning Theory, which will be discussed in what follows. We’ll see that it is naturally grouped with the somewhat older concept of a confidence interval. The character of Formal Learning Theory Formal Learning Theory is a collection of theorems about games of the following character. PLAYERS : You (the reader) and us (the authors). GAME PIECES : The set {0, 1, . . .} of natural numbers, denoted N. YOUR GOAL : Guess which subset of N we have in mind. ∗Contact: osherson@, weinstein@. 1 OUR GOAL : Pick a subset of N that you’ll fail to guess. RULES : First, all parties agree to a family C of nonempty subsets of N that are legal choices. We then pick nonempty S ∈ C and an ω-sequence e that orders S.1 We reveal e to you one member at a time. At stage n of this process (that is, once you’ve seen