Reinforcement learning (RL) is a powerful technique that allows agents to learn optimal decision-making policies through interactions with an environment. However, traditional RL algorithms suffer from several limitations such as the need for large amounts of data and long-term credit assignment, i.e. the problem of determining which actions actually produce a certain reward. Recently, Transformers have shown their capacity to address these constraints in this area of learning in an offline setting. This paper proposes a framework that uses Transformers to enhance the training of online off-policy RL agents and address the challenges described above through self-attention. The proposal introduces a hybrid agent with a mixed policy that combines an online off-policy agent with an offline Transformer agent using the Decision Transformer architecture. By sequentially exchanging the experience replay buffer between the agents, the agent's learning training efficiency is improved in the first iterations and so is the training of Transformer-based RL agents in situations with limited data availability or unknown environments.

One of the major problems of concern to the nowadays society is pollution, which can be of many types: acoustic, environmental, thermal, etc. Among these, noise pollution causes serious problems for citizens because it is continuous for a large part of the day, due to the fact that it is mostly caused by traffic. On the other hand, large cities provide a large amount of data obtained daily thanks to the sensorisation resulting from the concept of “smart cities”, which makes it possible to display information from the sensorised areas and to alert the institutions of the problems and, for citizens, to know the situation of noise pollution based on data in order to be able to make the relevant complaints and denunciations to the institutions. A universally understandable way of displaying the information contained in the captured data is the generation of linguistic descriptions that synthesise the information residing in the data. This paper presents a method for generating linguistic descriptions based on the noise pollution data captured by noise measurement stations. A method for generating descriptions of a day will be presented that considers the daily periods in which the data taken from the stations are structured (daytime, evening, night-time and full day). In order to test the proposed method, available data from the city of Madrid have been used to generate descriptions that allow the influence of Covid-19 on noise pollution to be analysed.

Abstract The backpropagation (BP) algorithm is a gradient-based algorithm used for training a feedforward neural network (FNN). Despite the fact that BP is still used today when FNNs are trained, it has some disadvantages, including the following: (i) it fails when non-differentiable functions are addressed, (ii) it can become trapped in local minima, and (iii) it has slow convergence. In order to solve some of these problems, metaheuristic algorithms have been used to train FNN. Although they have good exploration skills, they are not as good as gradient-based algorithms at exploitation tasks. The main contribution of this article lies in its application of novel memetic approaches based on the Gravitational Search Algorithm (GSA) and Chaotic Gravitational Search Algorithm (CGSA) algorithms, called respectively Memetic Gravitational Search Algorithm (MGSA) and Memetic Chaotic Gravitational Search Algorithm (MCGSA), to train FNNs in three classical benchmark problems: the XOR problem, the approximation of a continuous function, and classification tasks. The results show that both approaches constitute suitable alternatives for training FNNs, even improving on the performance of other state-of-the-art metaheuristic algorithms such as ParticleSwarm Optimization (PSO), the Genetic Algorithm (GA), the Adaptive Differential Evolution algorithm with Repaired crossover rate (Rcr-JADE), and the Covariance matrix learning and Bimodal distribution parameter setting Differential Evolution (COBIDE) algorithm. Swarm optimization, the genetic algorithm, the adaptive differential evolution algorithm with repaired crossover rate, and the covariance matrix learning and bimodal distribution parameter setting differential evolution algorithm.

Time series (TSs) are usually represented numerically; however, there are many situations where a linguistic description is preferable. Granular linguistic model of phenomena (GLMP) is a paradigm used in the generation of linguistic descriptions of static situations without considering the temporal relationship of data present, for example, in TSs. This paper presents a new method to generate linguistic descriptions of TSs with an operation similar to petri nets (PNs) and inspired by GLMP. The presented approach maintains the operation of PNs, adding a mechanism to generate linguistic descriptions based on GLMP. The main components of GLMP are added to places and transitions of PNs. This extension is called linguistic PNs (LPNs) and is a language that can be used to generate linguistic descriptions of systems. GLMP is a method that can be used by experts to design how the linguistic descriptions are synthesized and generated. So, LPNs also allow incorporating expert knowledge and combining descriptions in an appropriate way. The experimental part is focused on showing how LPNs can be used to linguistically describe TSs, including the occurrence of maxima or minima, and trends.

In the last few years, hybridization has spread as an effective technique to solve hard optimization problems where metaheuristics algorithms have been unable to find global optima in a computational cost given. In this article, we propose the so-called cooperation strategy. This way is an alternative to hybridization in which different algorithms work together in order to find a global optimum following an intrinsically parallel approach. Different homogeneous and heterogeneous strategies using CEC benchmark functions have been designed using Brain Storm Optimization (BSO) metaheuristic in comparison with a hybrid BSO, showing that cooperation improves significantly hybridization results and the original BSO.

This paper presents a system for detecting road departures by comparing linguistic representations for the trajectory of the vehicle with that for the lane marks of the road. All this information is obtained from a single camera processing exclusively the H264 motion vectors extracted from the recorded video. The process of comparison between the linguistic elements allows detecting the subset of continuous frames where there is no logical correspondence between the displacement of the vehicle and the road shape. Since the videos are captured from a moving vehicle, we propose a statistically based process to use domain changing fuzzy sets adapted to traffic scenarios that continuously change. This improves the reliability of the linguistic descriptions that, once compared, are used to detect departures. Lastly, a set of experiments using traffic videos with different characteristics are presented to validate this approach.

This paper tries the use of the fuzzy logic to detect olive trees in Very High Resolution Images. More concretely, fuzzy numbers are used to manage the imprecise information of the input images. A method to summarize information in a fuzzy number from data is used, and after that, the obtained fuzzy number is used to process the Very High Resolution Images by using a k-neighbor strategy. In this first approach, tested images show encouraging results because all olive trees are detected.

Intelligent surveillance refers to using Artificial Intelligence techniques in order to improve surveillance and deal with semantic information obtained from low-level security devices. In this context, the use of expert knowledge may offer a more realistic solution when dealing with the design of a surveillance system. In this work, a conceptual framework based on normality analysis to detect abnormal behaviors by means of normality concepts is presented. A normality concept specifies how a certain object should ideally behave in a concrete environment depending on such a concept. The definition of the normal path concept is studied in depth in order to analyze behaviors in an outdoor environment.

In this paper, we propose the use of temporal fuzzy chains for the modeling of dynamical systems in a way that is comprehensible. We are interested in helping the overall understanding of the system execution, over and during a precise and finite time. To this end, we model its input/output behavior and how this has changed in the past. There is a double goal in mind: accuracy and interpretability. An inductive algorithm for analyzing finite continuous multivariate time series will be achieved, in which the use of fuzzy logic has been taken into account. The aim of the algorithm is to help us to find changes in a system, as well as to identify the causes of these changes in a linguistic form. The causes will be specified by means of a set of fuzzy transitions between consecutive states, which consist of fuzzy rules that model the system. The method suggested has been applied on a real life case, human walk modeling.

Temporal fuzzy models (TFMs) are used to represent the dynamic systems (DSs) in a linguistic way. TFMs are composed of ordered temporal fuzzy rules (TFRs), thus, each rule represents a temporal zone. Normally, we obtain the TFMs by a fuzzy model induced by means of another induction algorithm, that is: first of all, an induction algorithm is used to obtain a fuzzy model, and secondly, an algorithm is used to transform the obtained fuzzy model into a TFM. The aim of this paper is to present a direct induction algorithm of TFMs, that is, to present a method capable of obtaining a TFM which represents the DS study through the use of a time series as an input.

In this work, we present the induction of a fuzzy model that represents the behavior of a partial known function. We extend the approach of classical induction of a classifier by building a decision tree, and its generalization for regression problems by CART, to build a fuzzy model. It is defined by a collection of fuzzy regions fixed in the input domain of the function. To obtain fuzzy regions of the input domain, we have defined a new method (FCMD) to get a fuzzy partition of a fuzzy set, which generalizes the classical Bezdeck's method FCM. © 2001 John Wiley & Sons, Inc.

It is generally accepted in the information system (IS) field that IS quality is highly dependent on the decisions made early in the development life cycle. The construction of conceptual data models is often an important task of this early development. Therefore, improving the quality of conceptual data models will be a major step towards the quality improvement of the IS development. Several quality frameworks for conceptual data models have been proposed, but most of them lack valid quantitative measures in order to evaluate the quality of conceptual data models in an objective way. In this article we will define measures for the structural complexity (internal attribute) of entity relationship diagrams (ERD) and use them for predicting their maintainability (external attribute). We will theoretically validate the proposed metrics following Briand et al.'s framework with the goal of demonstrating the properties that characterise each metric. We will also show how it is possible to predict each of the maintainability sub-characteristics using a prediction model generated using a novel method for induction of fuzzy rules.