Vovk, Vladimir (2006) Leading strategies in competitive on-line prediction.
Full text access: Open
We start from a simple asymptotic result for the problem of on-line regression with the quadratic loss function: the class of continuous limited-memory prediction strategies admits a "leading prediction strategy", which not only asymptotically performs at least as well as any continuous limited-memory strategy but also satisfies the property that the excess loss of any continuous limited-memory strategy is determined by how closely it imitates the leading strategy. More specifically, for any class of prediction strategies constituting a reproducing kernel Hilbert space we construct a leading strategy, in the sense that the loss of any prediction strategy whose norm isnot too large is determined by how closely it imitates the leading strategy. This result is extended to the loss functions given by Bregman divergences and by strictly proper scoring rules.
This is a Submitted version This version's date is: 27/7/2006 This item is not peer reviewed
https://repository.royalholloway.ac.uk/items/fa3b7c25-d01a-f0cf-3576-b5ec85d4a16f/7/
Deposited by Research Information System (atira) on 22-Jul-2014 in Royal Holloway Research Online.Last modified on 22-Jul-2014
20 pages; a conference version published in the ALT 2006 proceedings