On calibration error of randomized forecasting algorithms

Vladimir V. V'yugin

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

Abstract

Recently, it was shown that calibration with an error less than δ > 0 is almost surely guaranteed with a randomized forecasting algorithm, where forecasts are chosen using randomized rounding up to S of deterministic forecasts. We show that this error can not be improved for a large majority of sequences generated by a probabilistic algorithm: we prove that combining outcomes of coin-tossing and a transducer algorithm, it is possible to effectively generate with probability close to one a sequence "resistant" to any randomized rounding forecasting with an error much smaller than δ.

Original languageEnglish
Title of host publicationAlgorithmic Learning Theory - 18th International Conference, ALT 2007, Proceedings
PublisherSpringer Verlag
Pages388-402
Number of pages15
ISBN (Print)9783540752240
DOIs
Publication statusPublished - 2007
Externally publishedYes
Event18th International Conference on Algorithmic Learning Theory, ALT 2007 - Sendai, Japan
Duration: 1 Oct 20074 Oct 2007

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume4754 LNAI
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Conference

Conference18th International Conference on Algorithmic Learning Theory, ALT 2007
Country/TerritoryJapan
CitySendai
Period1/10/074/10/07

Fingerprint

Dive into the research topics of 'On calibration error of randomized forecasting algorithms'. Together they form a unique fingerprint.

Cite this