The support vector machine (SVM) with Gaussian kernel often achieves state-of-the-art performance in classification problems, but requires the tuning of the kernel spread. Most optimization methods for spread tuning require training, being slow and not suited for large-scale datasets. We formulate an analytic expression to calculate, directly from data without iterative search, the spread minimizing the difference between Gaussian and ideal kernel matrices. The proposed direct gamma tuning (DGT) equals the performance of and is one to two orders of magnitude faster than the state-of-the art approaches on 30 small datasets. Combined with random sampling of training patterns, it also runs on large classification problems. Our method is very efficient in experiments with 20 large datasets up to 31 million of patterns, it is faster and performs significantly better than linear SVM, and it is also faster than iterative minimization. Code is available upon paper acceptance from this link: https://persoal.citius.usc.es/manuel.fernandez.delgado/papers/dgt/index.html and from CodeOcean: https://codeocean.com/capsule/4271163/tree/v1.

Abstract Federated learning is the state-of-the-art paradigm for training a learning model collaboratively across multiple distributed devices while ensuring data privacy. Under this framework, different algorithms have been developed in recent years and have been successfully applied to real use cases. The vast majority of work in federated learning assumes static datasets and relies on the use of deep neural networks. However, in real-world problems, it is common to have a continual data stream, which may be non-stationary, leading to phenomena such as concept drift. Besides, there are many multi-device applications where other, non-deep strategies are more suitable, due to their simplicity, explainability, or generalizability, among other reasons. In this paper we present Ensemble and Continual Federated Learning, a federated architecture based on ensemble techniques for solving continual classification tasks. We propose the global federated model to be an ensemble, consisting of several independent learners, which are locally trained. Thus, we enable a flexible aggregation of heterogeneous client models, which may differ in size, structure, or even algorithmic family. This ensemble-based approach, together with drift detection and adaptation mechanisms, also allows for continual adaptation in situations where data distribution changes over time. In order to test our proposal and illustrate how it works, we have evaluated it in different tasks related to human activity recognition using smartphones.

Abstract Smart devices, such as smartphones, wearables, robots, and others, can collect vast amounts of data from their environment. This data is suitable for training machine learning models, which can significantly improve their behavior, and therefore, the user experience. Federated learning is a young and popular framework that allows multiple distributed devices to train deep learning models collaboratively while preserving data privacy. Nevertheless, this approach may not be optimal for scenarios where data distribution is non-identical among the participants or changes over time, causing what is known asconcept drift. Little research has yet been done in this field, but this kind of situation is quite frequent in real life and poses new challenges to both continual and federated learning. Therefore, in this work, we present a new method, called Concept-Drift-Aware Federated Averaging (CDA-FedAvg). Our proposal is an extension of the most popular federated algorithm, Federated Averaging (FedAvg), enhancing it for continual adaptation under concept drift. We empirically demonstrate the weaknesses of regular FedAvg and prove that CDA-FedAvg outperforms it in this type of scenario.

The support vector machine (SVM) is a very important machine learning algorithm with state-of-the-art performance on many classification problems. However, on large datasets it is very slow and requires much memory. To solve this defficiency, we propose the fast support vector classifier (FSVC) that includes: 1) an efficient closed-form training free of any numerical iterative procedure; 2) a small collection of class prototypes that avoids to store in memory an excessive number of support vectors; and 3) a fast method that selects the spread of the radial basis function kernel directly from data, without classifier execution nor iterative hyper-parameter tuning. The memory requirements of FSVC are very low, spending in average only 6$\cdot 10^{-7}$sec. per pattern, input and class, and processing datasets up to 31 millions of patterns, 30,000 inputs and 131 classes in less than 1.5 hours (less than 3 hours with only 2GB of RAM). In average, the FSVC is 10 times faster, requires 12 times less memory and achieves 4.7 percent more performance than Liblinear, that fails on the 4 largest datasets by lack of memory, being 100 times faster and achieving only 6.7 percent less performance than Libsvm. The time spent by FSVC only depends on the dataset size and thus it can be accurately estimated for new datasets, while Libsvm or Liblinear are much slower on “difficult” datasets, even if they are small. The FSVC adjusts its requirements to the available memory, classifying large datasets in computers with limited memory. Code for the proposed algorithm in the Octave scientific programming language is provided.1

The SoftLearn Activity Reporter is a data-to-text service, which automatically generates textual reports about the activity developed by students within the SoftLearn virtual learning environment. In this paper, we describe the conception of the service, its architecture, and its subsequent evaluation by an expert pedagogue, where 20 full reports generated from real data from an undergraduate course supported by the SoftLearn platform were assessed. Results show that the automatically generated reports are a valuable complementary tool for explaining teachers and students the information comprised in a learning analytics dashboard.

In this paper we present a general model for building linguistic descriptions of data (LDD) solutions, which is based on computational models of perception inspired in the computational theory of perceptions (CTP) and in fields of knowledge different from the computational intelligence area. The elements in the model aim to consider the richness and complexity that real LDD processes are endowed with and their actual role in data-to-text natural language generation (D2T-NLG) systems. As an example of application, the building of a LDD solution in the meteorology realm is presented.

In this paper we present the Soft Learn Activity Reporter (SLAR) service which automatically generates textual short-term reports about learners' behavior in virtual learning environments. Through this approach, we show how textual reporting is a coherent way of providing information that can complement (and even enhance) visual statistics and help teachers to understand in a comprehensible manner the behavior of their students during the course. This solution extracts relevant information from the students' activity and encodes it into intermediate descriptions using linguistic variables and temporal references, which are subsequently translated into texts in natural language. The examples of application on real data from an undergraduate course supported by the Soft Learn platform show that automatic textual reporting is a valuable complementary tool for explaining teachers and learners the information comprised in a Learning Analytics Dashboard.

We present in this paper an application that automatically generates textual short-term weather forecasts for every municipality in Galicia (NW Spain), using the real data provided by the Galician Meteorology Agency (MeteoGalicia). This solution combines in an innovative way computing with perceptions techniques and strategies for linguistic description of data, together with a natural language generation (NLG) system. The application, which is named GALiWeather, extracts relevant information from weather forecast input data and encodes it into intermediate descriptions using linguistic variables and temporal references. These descriptions are later translated into natural language texts by the NLG system. The obtained forecast results have been thoroughly validated by an expert meteorologist from MeteoGalicia using a quality assessment methodology, which covers two key dimensions of a text: the accuracy of its content and the correctness of its form. Following this validation, GALiWeather will be released as a real service, offering custom forecasts for a wide public.

Learning objects have arisen in response to the need of high-quality and reusable instructional materials. The repositories that hold learning objects allow educators to create and share their instructional contents in an organized infrastructure and where information should be easily searchable. Therefore, a key point of these repositories is the way learning objects are categorized. In this paper, we present an approach for classifying semantically learning objects whose metadata are described in Dublin Core. Specifically, our objective is to annotate automatically the subject of the learning object with instances of the DBpedia ontology, that is, to annotate the learning objects repository with linked open data. The syntactic and semantic analysis of the learning object will drive the selection of the most appropriate categories in the linked data repository.

Parallel perceptrons (PPs) are very simple and efficient committee machines (a single layer of perceptrons with threshold activation functions and binary outputs, and a majority voting decision scheme), which nevertheless behave as universal approximators. The parallel delta (P-Delta) rule is an effective training algorithm, which, following the ideas of statistical learning theory used by the support vector machine (SVM), raises its generalization ability by maximizing the difference between the perceptron activations for the training patterns and the activation threshold (which corresponds to the separating hyperplane). In this paper, we propose an analytical closed-form expression to calculate the PPs' weights for classification tasks. Our method, called Direct Parallel Perceptrons (DPPs), directly calculates (without iterations) the weights using the training patterns and their desired outputs, without any search or numeric function optimization. The calculated weights globally minimize an error function which simultaneously takes into account the training error and the classification margin. Given its analytical and noniterative nature, DPPs are computationally much more efficient than other related approaches (P-Delta and SVM), and its computational complexity is linear in the input dimensionality. Therefore, DPPs are very appealing, in terms of time complexity and memory consumption, and are very easy to use for high-dimensional classification tasks. On real benchmark datasets with two and multiple classes, DPPs are competitive with SVM and other approaches but they also allow online learning and, as opposed to most of them, have no tunable parameters.

We propose a method, called Direct Kernel Perceptron (DKP), to directly calculate the weights of a single perceptron using a closed-form expression which does not require any training stage. The weigths minimize a performance measure which simultaneously takes into account the training error and the classification margin of the perceptron. The ability to learn non-linearly separable problems is provided by a kernel mapping between the input and the hidden space. Using Gaussian kernels, DKP achieves better results than the standard Support Vector Machine (SVM) and Linear Discriminant Analysis (LDA) for a wide variety of benchmark two-class data sets. The computational cost of DKP linearly increases with the dimension of the input space and it is much lower than the corresponding to SVM.

The Parallel Perceptron (PP) is a simple neural network which has been shown to be a universal approximator, and it can be trained using the Parallel Delta (P-Delta) rule. This rule tries to maximize the distance between the perceptron activations and their decision hyperplanes in order to increase its generalization ability, following the principles of the Statistical Learning Theory. In this paper we propose a closed-form analytical expression to calculate, without iterations, the PP weights for classification tasks. The calculated weights globally optimize a cost function which takes simultaneously into account the training error and the perceptron margin, similarly to the P-Delta rule. Our approach, called Direct Parallel Perceptron (DPP) has a linear computational complexity in the number of inputs, being very interesting for high-dimensional problems. DPP is competitive with SVM and other approaches (included P-Delta) for two-class classification problems but, as opposed to most of them, the tunable parameters of DPP do not influence the results very much. Besides, the absence of an iterative training stage gives to DPP the ability of on-line learning.

This paper proposes polytope ARTMAP (PTAM), an adaptive resonance theory (ART) network for classification tasks which does not use the vigilance parameter. This feature is due to the geometry of categories in PTAM, which are irregular polytopes whose borders approximate the borders among the output predictions. During training, the categories expand only towards the input pattern without category overlap. The category expansion in PTAM is naturally limited by the other categories, and not by the category size, so the vigilance is not necessary. PTAM works in a fully automatic way for pattern classification tasks, without any parameter tuning, so it is easier to employ for nonexpert users than other classifiers. PTAM achieves lower error than the leading ART networks on a complete collection of benchmark data sets, except for noisy data, without any parameter optimization.

ART neural networks are important tools for online supervised pattern recognition. They use internal categories with pre-defined geometry, given by the category choice function. Pre-defined geometry limits the ability of the categories to fit complex borders among output predictions for a given data set, and may contribute to the category proliferation problem. This work proposes Polytope ARTMAP (PTAM), whose category representation regions have general geometry-polytopes in R/sup n/ whose vertices are selected training patterns. The category borders compose a piece-wise linear approximation to the borders among predictions. Overlapping among categories is avoided in PTAM because they do not need to overlap in order to keep their geometry during learning. The choice function does not depend on the category size. Category growing is only limited by the other categories, and the vigilance parameter can be removed, so that PTAM learns a training data set without any parameter tuning.

In this paper, we propose a new quantifier fuzzification mechanism which is deeply rooted in the theory of probability. This quantifier fuzzification mechanism skips the nested assumption, which is inherent to other probabilistic quantification methods. The new quantification approach complies with the properties required for determiner fuzzification schemes (DFS) with finite sets and, hence, its good behavior is assured. Moreover, this new approach is suitable for some application domains. In particular, the use of fuzzy quantifiers for implementing query quantified statements for information retrieval exemplifies the adequacy of the new proposal. The new quantifier fuzzification mechanism has been efficiently implemented and empirically tested for a retrieval task. This practical evaluation followed the standard methodology in the field of information retrieval and was conducted against a popular benchmark consisting of a large collection of documents. The retrieval performance evaluation made evident that: 1) the new method can work in realistic scenarios, and 2) it can overcome recent proposals for applying fuzzy quantifiers in information retrieval

Detection of landmarks is essential in mobile robotics for navigation tasks like building topological maps or robot localization. Doors are one of the most common landmarks since they show the topological structure of indoor environments. In this paper, the novel paradigm of fuzzy temporal rules is used for detecting doors from the information of ultrasound sensors. This paradigm can be used both to model the necessary knowledge for detection and to consider the temporal variation of several sensor signals. Experimental results using a Nomad 200 mobile robot in a real environment produce 91% of doors were correctly detected, which show the reliability and robustness of the system.

This paper describes an Integral Technical Information Management System (ITIMS), which has been designed to operate in industrial fields where it is necessary to monitor a number of processes. ITIMS is a distributed instrumentation environment with ubiquitous access to information that has been developed using standard information and communications technologies, and which is currently operative in the ENDESA-As Pontes coal-fired power station, located in Galicia (NW Spain).

This work describes the design and implementation of a knowledge based system that automatizes the furniture budgeting task. The aim of this task is to estimate the production cost that would suppose a furniture order manufacturing. Fulfillment of this aim requires the management of the company resources, information and knowledge in order to determine workload and financial impacts of client request manufacturing. Task modelling has been done through the workflows formalism approach and describes working activities and its relationships with the company resources. The system software architecture is based on workflow modeling and is an adjustment of a conceptual architecture proposed to be a standard for workflows systems.

In this paper, we propose the multivariable fuzzy temporal profile model (MFTP), which enables the projection of expert knowledge on a physical system over a computable description. This description may be used to perform automatic abstraction on a set of parameters that represent the temporal evolution of the system. This model is based on the constraint satisfaction problem (CSP) formalism, which enables an explicit representation of the knowledge, and on fuzzy set theory, from which it inherits the ability to model the imprecision and uncertainty that are characteristics of human knowledge vagueness. We also present an application of the MFTP model to the recognition of landmarks in mobile robotics, specifically to the detection of doors on ultrasound sensor signals from a Nomad 200 robot.

A framework for description of fuzzy quantification models is presented. Within this framework, the fuzzy quantified statements evaluation problem is described as the compatibility between the fuzzy quantifier and a fuzzy cardinality or a fuzzy aggregation measure. A list of desirable properties for quantification models is presented and those models that fit the framework are confronted with it.

Two approaches are presented for the modelling of fuzzy operators on fuzzy domains. The approaches are based on crisp definitions that operate onto crisp representatives of the involved fuzzy sets, which are obtained by using bivalued and trivalued level-cuts. Application examples in the field of fuzzy temporal rules-based reasoning are presented.

The paper describes a fuzzy control system for the avoidance of moving objects by a robot. The objects move with no type of restriction, varying their velocity and making turns. Due to the complex nature of this movement, it is necessary to realize temporal reasoning with the aim of estimating the trend of the moving object. A new paradigm of fuzzy temporal reasoning, which we call fuzzy temporal rules (FTRs), is used for this control task. The control system has over 117 rules, which reflects the complexity of the problem to be tackled. The controller has been subjected to an exhaustive validation process and examples are shown of the results obtained.

In this work we present the design proposal for a therapeutic specialist TS in the domain of the intensive coronary care unit ICCU for situations of acute myocardial infarct AMI . This TS can be placed within a more general framework of the intelligent supervision of patients interned in an ICCU, and it carries out two basic functions: 1 recommending to the physician that therapy should be administered according to the state and response of the treatment in progress; and 2 to continually monitor the evolution of the patient with the therapy in progress, advising the physician of necessary therapeutic modifications associations with other drugs, suspension of treatment, etc. in order to compensate for the evolution of the patient and aid his recovery. Thus we present a specialist agent architecture designed to carry out both functions, which is based on the identification, design, and implementation of therapeutic tasks, on the explicit separation between knowledge of the domain and the metaknowledge necessary in order for the coordinated accomplishment of these tasks, and on mimicking the methodology followed by the physician when evaluating the monitoring of the patient's condition.

In this paper, an enhanced model of fuzzy temporal knowledge bases (FTKBs) formed by fuzzy temporal rules is introduced, and its projection onto the formalism of fuzzy Petri nets (FPNs) is presented. The main objective is to enhance the expressiveness of the rules making up the FTKB, regarding a previous proposal,at the same time as keeping the computation cost within the limits that allow us to work on-line in dynamic environments. This is attained by means of a partial compaction of the rules, prior to their on-line execution. The projection of the FTKB over an FPN makes its design, analysis, and execution easier, by explicitly bringing out all its variables and relationships. © 1999 John Wiley & Sons, Inc.

We take as a setting point the Fuzzy Temporal Profile (FTP) model, which is used to describe the temporal evolution of a certain physical parameter. This model defines the approximation of an evolution curve by means of a set of linear sections between a series of significant points (X0, X1, …, XN), and in which each section is defined by means of an imprecise constraint on the duration, the increase in value and the slope between the points connected by the section. Following the FTP model we present a language for the representation of trends, the semantics of which is projected onto a fuzzy definition of each one of the constraints of a profile of interest.

This paper describes MART, an ART-based neural network for adaptive classification of multichannel signal patterns without prior supervised learning. Like other ART-based classifiers, MART is especially suitable for situations in which not even the number of pattern categories to be distinguished is known a priori; its novelty lies in its truly multichannel orientation, especially its ability to quantify and take into account during pattern classification the different changing reliability of the individual signal channels. The extent to which this ability can reduce the creation of spurious or duplicate categories (a major problem for ART-based classifiers of noisy signals) is illustrated by evaluation of its performance in classifying QRS complexes in two-channel ECG traces which were taken from the MIT-BIH database and contaminated with noise.

In this paper, we consider the problem of executing a fuzzy knowledge base (FKB) with rule chaining. The inference process used as starting point is the one based on forward reasoning functions which, obtained from the compositional rule of inference, permits performing the execution of rules in the truth space. This way the process is totally independent from the universes of discourse in which the different variables are defined, allowing a homogeneous treatment for all the variables in the FKB. The execution of the rules is interpreted as the "propagation" of linguistic truth values of the linguistic truth variable that reflect the linguistic degree of fulfillment of each of the propositions in the rules. This execution process is analyzed in two fields of application: control systems, where it is customary to assume t-norm operators as implication functions, and the aggregation process is implemented through the maximum operator and expert systems applications, where other implication functions may be needed and t-norm operators are generally used as aggregation operators. For both of these situations, we present a compaction mechanism which allows a noticeable part of the operations to be performed a priori, thus achieving an important computation time saving.

We develop a representational model for the knowledge base (KB) of fuzzy production systems with rule chaining based on the Petri net formalism. The model presents the execution of a KB following a data driven strategy based on the sup-min compositional rule of inference. In this connection, algorithms characterizing different situations have been described, including the case where the KB is characterized by complete information about all the input variables and the case where it is characterized by ignorance of some of these variables. For this last situation we develop a process of "incremental reasoning"; this process allows the KB to take information about previously unknown values into consideration as soon as such information becomes available. Furthermore, as compared to other solutions, the rule chaining mechanism we introduce is more flexible, and the description of the rules more generic. The computational complexity of these algorithms is O((C/2+M+N)R/sup 2/) for the "complete information" case and O((M+N)R/sup 2/) and O(2(M+N)R/sup 2/) for the other cases, where R is the number of fuzzy conditional statements of the KB, M and N the maximum number of antecedents and consequents in the rules and C the number of chaining transitions in the KB representation.>