Cerami M.,Palacky University |
Garcia-Cerdana A.,Institute dInvestigacio en IntelLigencia Artificial IIIA CSIC |
Garcia-Cerdana A.,University Pompeu Fabra |
Esteva F.,Institute dInvestigacio en IntelLigencia Artificial IIIA CSIC
International Journal of Approximate Reasoning | Year: 2014
This paper deals with finitely-valued fuzzy description languages from a logical point of view. From recent results in Mathematical Fuzzy Logic and following , we develop a Fuzzy Description Logic based on the fuzzy logic of a finite BL-chain. The constructors of the languages presented in this paper correspond to the connectives of that logic (containing an involutive negation, Monteiro-Baaz delta and hedges). The paper addresses the hierarchy of fuzzy attributive languages; knowledge bases and their reductions; reasoning tasks; and complexity. Our results regarding decidability together with a summary of the known results related to computational complexity are of particular interest. In Appendix B we also provide axiomatizations for expansions of the logic of a finite BL-chain considered in the paper. © 2013 Elsevier Inc.
Pardo P.,Institute dInvestigacio en Intelligencia Artificial IIIA CSIC |
Godo L.,Institute dInvestigacio en Intelligencia Artificial IIIA CSIC
Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) | Year: 2011
The aim of this paper is to offer an argumentation-based defeasible logic that enables temporal forward reasoning. We extend the DeLP logical framework by associating temporal parameters to literals. A temporal logic program is a set of temporal literals and durative rules. These temporal facts and rules combine into durative arguments representing temporal processes, that permit us to reason defeasibly about future states. The corresponding notion of logical consequence, or warrant, is defined slightly different from that of DeLP, due to the temporal aspects. As usual, this notion takes care of inconsistencies, and in particular we prove the consistency of any logical program whose strict part is consistent. Finally, we define and study a sub-class of arguments that seem appropriate to reason with natural processes, and suggest a modification to the framework that is equivalent to restricting the logic to this class of arguments. © 2011 Springer-Verlag.
Armengol E.,Institute DInvestigacio en Intelligencia Artificial IIIA CSIC
Frontiers in Artificial Intelligence and Applications | Year: 2013
Self-Training methods are a family of methods that uses some supervised method (commonly an inductive learning method) to assign class labels to unlabeled examples. The resulting inductive model is useful to predict the classification of unseen new domain objects. In this paper we propose to use a lazy learning method called LID, capable of producing descriptions similar to the ones from inductive learning methods. In the experiments we prove that this partial domain is very useful to predict the classification of unseen objects. © 2013 The authors and IOS Press. All rights reserved.