Linear Probability Forecasting

Abstract : In this paper we consider two online multi-class classification problems: classification with linear models and with kernelized models. The predictions can be thought of as probability distributions. The quality of predictions is measured by the Brier loss function. We suggest two computationally efficient algorithms to work with these problems, the second algorithm is derived by considering a new class of linear prediction models. We prove theoretical guarantees on the cumulative losses of the algorithms. We kernelize one of the algorithms and prove theoretical guarantees on the loss of the kernelized version. We perform experiments and compare our algorithms with logistic regression.
Type de document :
Communication dans un congrès
Harris Papadopoulos; Andreas S. Andreou; Max Bramer. 6th IFIP WG 12.5 International Conference on Artificial Intelligence Applications and Innovations (AIAI), Oct 2010, Larnaca, Cyprus. Springer, IFIP Advances in Information and Communication Technology, AICT-339, pp.4-11, 2010, Artificial Intelligence Applications and Innovations. 〈10.1007/978-3-642-16239-8_4〉
Liste complète des métadonnées

Littérature citée [10 références]  Voir  Masquer  Télécharger

https://hal.inria.fr/hal-01060645
Contributeur : Hal Ifip <>
Soumis le : jeudi 16 novembre 2017 - 15:56:04
Dernière modification le : dimanche 17 décembre 2017 - 01:11:24
Document(s) archivé(s) le : samedi 17 février 2018 - 15:46:41

Fichier

ZhdanovK10.pdf
Fichiers produits par l'(les) auteur(s)

Licence


Distributed under a Creative Commons Paternité 4.0 International License

Identifiants

Citation

Fedor Zhdanov, Yuri Kalnishkan. Linear Probability Forecasting. Harris Papadopoulos; Andreas S. Andreou; Max Bramer. 6th IFIP WG 12.5 International Conference on Artificial Intelligence Applications and Innovations (AIAI), Oct 2010, Larnaca, Cyprus. Springer, IFIP Advances in Information and Communication Technology, AICT-339, pp.4-11, 2010, Artificial Intelligence Applications and Innovations. 〈10.1007/978-3-642-16239-8_4〉. 〈hal-01060645〉

Partager

Métriques

Consultations de la notice

96

Téléchargements de fichiers

8