Presentation 1996/10/31
An On-line Prediction Algorithm Combining the Several Prediction Strategies
Ichiro TAJIKA, Eiji TAKIMOTO, Akira MARUOKA,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) Cesa-Bianchi et al, proposed an on-line prediction model, in which an algorithm predicts a binary value by combining the predictions of several prediction strategies. The loss of the algorithm is measured by the expected number of mistakes made on the bit sequence, and the goal of the algorithm is to minimize the difference between the loss it makes and the loss made by the best strategy on this sequence. In this paper, the model is extended to incorporate the notion of investment. In particular, the strategies and the algorithm are required to bet money in each prediction and the losses are measured by the total amount, of money lost. An optimal prediction algorithm depending on the strategies and an almost optimal one not depending on the strategies are given under the condition that all the strategies bet the same amount of money in a prediction.
Keyword(in Japanese) (See Japanese page)
Keyword(in English) on-line prediction model / prediction strategy / minimax strategy / weighted majority algorithm
Paper # COMP96-33
Date of Issue

Conference Information
Committee COMP
Conference Date 1996/10/31(1days)
Place (in Japanese) (See Japanese page)
Place (in English)
Topics (in Japanese) (See Japanese page)
Topics (in English)
Chair
Vice Chair
Secretary
Assistant

Paper Information
Registration To Theoretical Foundations of Computing (COMP)
Language JPN
Title (in Japanese) (See Japanese page)
Sub Title (in Japanese) (See Japanese page)
Title (in English) An On-line Prediction Algorithm Combining the Several Prediction Strategies
Sub Title (in English)
Keyword(1) on-line prediction model
Keyword(2) prediction strategy
Keyword(3) minimax strategy
Keyword(4) weighted majority algorithm
1st Author's Name Ichiro TAJIKA
1st Author's Affiliation Graduate School of Information Sciences, Tohoku University()
2nd Author's Name Eiji TAKIMOTO
2nd Author's Affiliation Graduate School of Information Sciences, Tohoku University
3rd Author's Name Akira MARUOKA
3rd Author's Affiliation Graduate School of Information Sciences, Tohoku University
Date 1996/10/31
Paper # COMP96-33
Volume (vol) vol.96
Number (no) 343
Page pp.pp.-
#Pages 10
Date of Issue