José Ranilla

Introducing a family of linear measures for feature selection in text categorization

Text categorization, which consists of automatically assigning documents to a set of categories, usually involves the management of a huge number of features. Most of them are irrelevant and others introduce noise which could mislead the classifiers. Thus, feature reduction is often performed in order to increase the efficiency and effectiveness of the classification. In this paper, we propose to select relevant features by means of a family of linear filtering measures which are simpler than the usual measures applied for this purpose. We carry out experiments over two different corpora and find that the proposed measures perform better than the existing ones.

F AN

In this paper we present a machine-learning algorithm that computes a small set of accurate and interpretable rules. The decisions of these rules can be straight-forwardly explained as the conclusions drawn by a case-based reasoner. Our system is named F AN, an acronym for f inding a ccurate i n ductions. It starts from a collection of training examples and produces propositional rules able to classify unseen cases following a minimum-distance criterion in their evaluation procedure. In this way, we combine the advantages of instance-based algorithms and the conciseness of rule (or decision-tree) inducers. The algorithm followed by F AN can be seen as the result of successive steps of pruning heuristics. The main tool employed is that of the impurity level, a measure of the classification quality of a rule, inspired by a similar measure used in IB3. Finally, a number of experiments were conducted with standard benchmark datasets of the UCI repository to test the performance of our system, successfully comparing F AN with a wide collection of machine-learning algorithms.

High-performance computing: the essential tool and the essential challenge

High-performance computing (HPC) is nowadays an essential tool for the solution of many problems that arise in both scientific and engineering realms. HPC platforms are based on clusters of multicore nodes, and half of these facilities all around the world also include some type of accelerator device such as graphics processing units (GPUs) or the Intel Xeon Phi coprocessor. Many research interests are addressed to optimize applications that can get the most of these configurations. At the same time, research on the HPC ecosystem (hardware, software tools, applications, etc.) is in the spotlight. In particular, exascale computing is receiving a major interest. The White House Office of Science and Technology Policy highlights the importance of exascale computing for the maintenance of US leadership over the coming decades, and it is for this reason that the United States is doing strategic investments in HPC to meet increasing computing demands and emerging technological challenges. Current and future research faces the natural problems that arise when concurrent resources becomevery large: huge electrical consumption, heat dissipation, and probability of failure, among others. Many problems arise as long as we proceed in the way to developing exascale systems. One of them is the increase of failure rates. This special issue presents the

The usefulness of artificial intelligence techniques to assess subjective quality of products in the food industry

In this paper we advocate the application of Artificial Intelligence techniques to quality assessment of food products. Machine Learning algorithms can help us to: (a) extract operative human knowledge from a set of examples; (b) conclude interpretable rules for classifying samples regardless of the non-linearity of the human behaviour or process; and (c) help us to ascertain the degree of influence of each objective attribute of the assessed food on the final decision of an expert. We illustrate these topics with an example of how it is possible to clone the behaviour of bovine carcass classifiers, leading to possible further industrial applications.

Improving performance of text categorization by combining filtering and support vector machines

Text Categorization is the process of assigning documents to a set of previously fixed categories. A lot of research is going on with the goal of automating this time-consuming task. Several different algorithms have been applied, and Support Vector Machines (SVM) have shown very good results. In this report, we try to prove that a previous filtering of the words used by SVM in the classification can improve the overall performance. This hypothesis is systematically tested with three different measures of word relevance, on two different corpus (one of them considered in three different splits), and with both local and global vocabularies. The results show that filtering significantly improves the recall of the method, and that also has the effect of significantly improving the overall performance.

Artificial intelligence techniques point out differences in classification performance between light and standard bovine carcasses

The validity of the official SEUROP bovine carcass classification to grade light carcasses by means of three well reputed Artificial Intelligence algorithms has been tested to assess possible differences in the behavior of the classifiers in affecting the repeatability of grading. We used two training sets consisting of 65 and 162 examples respectively of light and standard carcass classifications, including up to 28 different attributes describing carcass conformation. We found that the behavior of the classifiers is different when they are dealing with a light or a standard carcass. Classifiers follow SEUROP rules more rigorously when they grade standard carcasses using attributes characterizing carcass profiles and muscular development. However, when they grade light carcasses, they include attributes characterizing body size or skeletal development. A reconsideration of the SEUROP classification system for light carcasses may be recommended to clarify and standardize this specific beef market in the European Union. In addition, since conformation of light and standard carcasses can be considered different traits, this could affect sire evaluation programs to improve carcass conformation scores from data from markets presenting a great variety of ages and weights of slaughtered animals.

Using artificial intelligence to design and implement a morphological assessment system in beef cattle

In this paper a methodology is developed to improve the design and implementation of a linear morphological system in beef cattle using artificial intelligence. The proposed process involves an iterative mechanism where type traits are successively defined and computationally represented using knowledge engineering methodologies, scored by a set of trained human experts and finally, analysed by means of four reputed machine learning algorithms. The results thus achieved serve as feed back to the next iteration in order to improve the accuracy and efficacy of the proposed assessment system. A sample of 260 conformation records of the Asturiana de los Valles beef cattle breed is shown to illustrate the methodology. Three sources of inconsistency were detected: (a) the existence of different interpretations of the trait’s definition, increasing the subjectivity of the assessment; (b) the narrow range of variation of some of the anatomical traits assessed; (c) the inclusion of some complex traits in the assessment system. In this sense, the reopening of the evaluated Asturiana de los Valles assessment system is recommended. In spite of the difficulty of collecting data from live animals, further implications of the artificial intelligence systems on morphological assessment are pointed out.

Interpretability of fuzzy association rules as means of discovering threats to privacy

This paper focuses on studying how data privacy could be preserved with fuzzy rule bases as interpretable as possible. These fuzzy rule bases are obtained from a data mining strategy based on building a decision tree. The antecedents of each rule produced by these systems contain information about the released variables (quasi-identifier), whereas the consequent contains information only about the protected variable. Experimental results show that fuzzy rules are generally simpler and easier to interpret than other approaches but the risk of disclosing does not increase.

Neville elimination on multi- and many-core systems: OpenMP, MPI and CUDA

This paper describes several parallel algorithmic variations of the Neville elimination. This elimination solves a system of linear equations making zeros in a matrix column by adding to each row an adequate multiple of the preceding one. The parallel algorithms are run and compared on different multi- and many-core platforms using parallel programming techniques as MPI, OpenMP and CUDA.

Measures of Rule Quality for Feature Selection in Text Categorization

Text Categorization is the process of assigning documents to a set of previously fixed categories. A lot of research is going on with the goal of automating this time-consuming task. Several different algorithms have been applied, and Support Vector Machines have shown very good results. In this paper we propose a new family of measures taken from the Machine Learning environment to apply them to feature reduction task. The experiments are performed on two different corpus (Reuters and Ohsumed). The results show that the new family of measures performs better than the traditional Information Theory measures.