Premium
Guest Editorial: Recent Trends in Intelligent Systems
Author(s) -
Gámez José A.,
Herrera Francisco,
Puerta José M.
Publication year - 2017
Publication title -
international journal of intelligent systems
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.291
H-Index - 87
eISSN - 1098-111X
pISSN - 0884-8173
DOI - 10.1002/int.21831
Subject(s) - library science , citation , computer science , humanities , artificial intelligence , art history , art
The 16th Conference of the Spanish Association for Artificial Intelligence (CAEPIA 2015), was held in Albacete (Spain), during November 9–12, 2015. CAEPIA is a biennial forum open to worldwide researchers to present and discuss their last scientific and technological advances in AI. All perspectives — theory, methodology, and applications — are welcome. As a clear example of the good health enjoyed by worldwide AI research and particularly in the Spanish AI community, 175 papers were submitted to CAEPIA together with its associated tracks and workshops. This special issue on “Recent Trends in Intelligent Systems” consists of a careful selection of five papers, selected among the contributions presented at CAEPIA’ 2015. All of them were substantially extended and went through a new and strict review process. We think the selection represents a clear picture of the recent trends in intelligent systems research done by the Spanish AI community. The first three papers fall in the field of machine learning while the last two are related to reasoning tasks: diagnosis and text generation. In the next paragraphs we briefly describe the five contributions. Hernández-González et al. describe a new weakly supervised problem, termed as learning from positive-unlabelled proportions. This is a non-standard classification problem in which instances are unlabelled, but some class information is available. In particular, the training set is divided into subsets and the proportion of positive and unlabelled examples for each subset is known. The goal is to learn a standard classifier able to label unknown instances as positive or negative. Methods are provided to learn Bayesian networks classifiers by using EM algorithm. In Ramı́rez-Gallego et al. the problem of supervised feature selection in a high dimensional big data context is approached. The well-known minimumRedundancy-Maximum-Relevance (mRMR) feature subset selector is taken as a starting point and a fast version (fast-mRMR) is designed in order to overcome the computational burden of mRMR when dealing with high dimensional datasets. A set of optimizations for the original mRMR algorithm are described and a software package containing the implementation of fast-mRMR for three different platforms (single-CPU, GPU and Apache Spark) is delivered.