< Terug naar vorige pagina

Publicatie

PU-learning disjunctive concepts in ILP

Tijdschriftbijdrage - Tijdschriftartikel Conferentiebijdrage

© by the paper's authors. PU learning is a variant of semi-supervised learning where labels of only one class are given. In this late-breaking paper, we present early results on an ILP method that PU-learns disjunctive concepts. The problem is motivated by some work on natural language learning, more specifically, learning the meaning of n-grams from (sentence, context)-pairs, where the context is described using first-order logic. An earlier solution for this task could learn only conjunctive concepts. We show experimentally that the novel method effectively learns disjunctive concepts from PU data. We also discuss challenges and limitations.
Tijdschrift: 27th ILP 2017: Orléans, France - Late Breaking Papers
ISSN: 1613-0073
Volume: 2085
Pagina's: 6 - 10
Jaar van publicatie:2017
Toegankelijkheid:Open