Metaheuristic algorithms are widely used for solving optimization problems, and their performance critically depends on parameter settings. The irace tool has been proposed as an efficient method for automatic parameter tuning, identifying effective configurations while managing computational resources. However, irace can suffer from premature convergence, of-ten evaluating similar parameter configurations. This work introduces an approach to enhance irace’s exploration using clustering algorithms. The strategy groups elite configurations based on their location in the parameter space and selects parents from these groups to guide the sampling of new configurations, extending irace’s exploratory behavior. The method was evaluated by tuning the Ant Colony Optimization framework in three scenarios: a homogeneous setting (TSP instances with 2000 cities), a slightly less homogeneous one (TSP instances with 1000–3000 cities), and a heterogeneous setting (QAP instances with two levels of sparsity). Results show that the proposed strategy improves exploration, reaching structurally diverse configurations, though with scenario-dependent performance trade-offs.
Buildings are crucial for achieving decarbonization by 2050. In 2022, they represented 34% of global energy demand and 37% of CO 2 emissions related to energy and processes. Consequently, there has been significant interest in evaluating building energy efficiency from various perspectives over the past decade. In this paper, machine learning (ML) methods based on a meta-approach known as ensemble learning are proposed to predict the energy consumption of a university building. For this purpose, multivariate time series and multi-step prediction horizons are considered in two scenarios with different timestep resolutions. Ensemble learning methods such as Bagging (Random Forest), Boosting (Extreme Gradient Boosting), and Stacking are analyzed. The results show that Extreme Gradient Boosting achieves the best prediction error metrics RMSE, MAE, and MAPE in both scenarios. These models enhance predictive control systems in buildings for accurate short-term energy consumption forecasts.
The exponential growth in the amount and diversity of data organizations generate has presented significant challenges in data storage, processing, and management. Data Lakehouses, a hybrid solution that combines the capabilities of Data Warehouses and Data Lakes, have been developed to address these challenges. However, metadata management in these environments is crucial to prevent data from becoming inaccessible and unwieldy. We propose a metadata management model for a Data Lakehouse based on a functional architecture that integrates functional metadata features for a Data Lake with business and maintenance metadata for a Data Warehouse. We include the results of an initial implementation and discuss its impact on data accessibility and governance.
The integration of Information and Communication Technologies (ICT) in higher education has seen continuous growth, offering numerous tools and platforms to enhance the learning process for both students and educators. However, the lack of pedagogical training among many teachers prevents the use of ICT from automatically guaranteeing quality education. This article introduces a collaborative platform designed to facilitate educators in creating learning modules. The platform considers three key elements when designing a module: i) assessment as the guiding axis for the design, ii) the SOLO taxonomy to guide the definition of the learning outcomes, and iii) a conversational framework to assist in designing the various tasks that students must complete. Although the platform showed positive results a few years ago, evaluation revealed several areas for improvement in terms of technical, pedagogical, and usability aspects. As a result, improvements were implemented on three levels. First, the user experience was optimized through significant enhancements in the interface and collaboration functionalities, making the platform more intuitive and accessible. Second, the pedagogical support was updated, replacing Bloom's taxonomy with the SOLO taxonomy, which provides a more comprehensive and detailed approach to assessing student understanding and learning outcomes. Finally, the technological implementation was enhanced using advanced software tools, allowing for more efficient and robust development of features. These improvements not only optimized the user experience but also provided stronger pedagogical support and more advanced technological implementation, ensuring that the platform remains a valuable and effective tool in higher education. This experience yielded satisfactory results, both in evaluating the platform's usage by educators and in assessing students' experiences when working with learning modules generated by the platform.
Zero trust architecture (ZTA) is a novel security approach to building secure systems. For ZTA-based systems, several quality attributes (QAs) are critical given the challenges presented by its principles such as continuous verification, least privilege, and network micro-segmentation. Although several studies have contributed to the analysis the quality of ZTA-based systems, a systematic understanding of their QAs and the scope of these according to a well-established quality model is still missing. This article presents the results of a systematic literature review to identify most reported QAs for ZTA-based systems and their coverage, according to the well-known ISO/IEC 25010 quality model. The review of 22 primary studies shows that (i) 17 QAs were identified, but security and performance are the most frequently reported; (ii) only 8 QAs are mentioned in ISO/IEC 25010; and (iii) ZTA-based systems have complex QAs interdependencies, like resiliency, that are only partially covered in ISO/IEC 25010. These preliminary findings on ZTA-based systems quality are essential for researchers and practitioners, offering theoretical and practical perspectives to enhance understanding and implementation of this novel security approach.
The digital transformation, accelerated by COVID-19, has impacted higher education, forcing universities to adapt training models, methods, and strategies to take advantage of technological resources and align students' skills with the new demands of the labor market. This article aims to recognize which strategies universities use to acquire those skills demanded by Industry 4.0 and which factors determine success in their implementation. We used an exploratory literature review protocol based on the PICO methodology, consulting the Scopus database. The results show the implementation of various strategies that use technological tools to develop emerging skills in teachers and students. Future work could consider the impact of disruptive technologies, such as the metaverse or artificial intelligence, on higher education or recognize what changes have been introduced in secondary education in the 4.0 era.
Information and Communication Technologies (ICTs) have been integrated into a variety of processes, with education being no exception. In this context, gamified environments emerge as a relevant resource that can be applied in diverse ways within the educational framework. This article explores the acceptance and implementation of a gamified setting in higher education through the application of the Technology Acceptance Model (TAM). It employs a sequential methodology for both the research conducted and the development of the gamified environment. As a result, the integration of gamified environments in higher education is demonstrated to be a valuable and engaging strategy that enhances the learning experience. Findings indicate that students exhibit enthusiasm for the educational process when gamification elements are incorporated, highlighting the model's potential for both pedagogical and technological applications.
Technology adoption is especially challenging in agricultural settings. Maturity models are commonly used to assess the technology adoption readiness in such settings. While they have been shown to be good enough assessment tools, they typically challenge people when using them because each country have particular situations. We are working towards developing a Chilean-context specific technology adoption readiness assessment model. We present a real-world case in which we retrospectively use our maturity model to assess two key domain areas: Human resources and Infrastructure (both part of Capacity management domain). Promising results presented in this paper show that the maturity model allows assessments to be more organized and focused. Ongoing work to improve this maturity model is underway. We expect that the model would eventually allow self-assessments in agricultural settings.
Climate change has had a significant impact on honey bee production. For instance, droughts affect the availability of bees' food resources. Consequently, one emerging line of research involves the application of machine learning to model honey production. This paper presents a bibliographic review of machine learning research aimed at predicting honey yield. A comprehensive literature review was conducted using the primary scientific database known as WoS, followed by the systematization of information. The most noteworthy models identified are MARS (Multivariate Adaptive Regression Splines) and GBR (Gradient Boosting Regression). When applying MARS, the variables used include survey records, questions pertaining to social, economic, educational aspects, honey bee races, and production records. This model achieved a determination coefficient (r) of 0.920. With the GBR model, the main variables used relate to climatic records, image use, and harvest information. This technique can predict with an average error of +/- 10.3 kg for weight, with this weight being in the correct class 82% of the time. Current studies primarily employ more traditional statistics and machine learning techniques, with few exploring more innovative methodologies such as deep neural networks. Implementing these advanced algorithms poses a challenge, further compounded by the limited availability of unified databases for these developments, with information scattered across multiple sources.
Skyline Query is a type of database query that filters out a set of interesting data points from a potentially large dataset based on multiple criteria. Until now, algorithms which tackle this query work over plain data or indexed data structures, but there does not exist a mechanism to address it in compact data. Nevertheless, this query can be addressed with the points indexed in Compact Data Structures. Specifically, we address two variants of the query in the k 2 -tree compact structure. One is related to a multi-criteria restriction named Constrained Skyline Query and the other counts the number of points dominated by every Skyline point which is called Enumerating Skyline. Our study confirms that the k 2 -tree compact data structure is a suitable structure to directly implement the Skyline Query. The experimentation of our proposal demonstrates a better use of resources than a query over non-indexed sets and can therefore be calculated without uncompressing the data.
Nonverbal behavior mimicry is fundamental for effective social interaction and group cohesion, especially in dynamic professional environments such as agile teams. However, the automatic detection and analysis of technique in these contexts present significant challenges due to the complexity of human interactions and the diversity of nonverbal expressions. This article presents the development of a predictive model for detecting nonverbal behavior mimicry in a collaborative contexts using Multimodal Learning Analytics, which collects and analyzes data from multiple sources. The results show advances in the automatic detection of mimicry in groups of more than two people, using a long-short term memory (LSTM) machine learning model focused on complete participant information. This innovative approach allows learning the progression of “storylines”, allowing the model to determine when to forget or integrate past events to inform decision-making. In this way, it is possible to achieve a better understanding of how episodes of mimicry occur during collaborative activities. To visualize these episodes, various visualizations are presented, illustrating the interactions and similarities in nonverbal behavior among team members. This approach not only facilitates the identification of mimicry patterns but also offers an alternative for understanding group dynamics in agile environments, thus favoring the efficiency and effectiveness of work teams.
Non-invasive and reliable methods are essential in the diagnostics and treatment planning of diabetic neuropathies. Forecasting models based on postural data seem to be a promising solution to this problem. However, the performance of machine learning models is often hindered by limited number of observations and imbalanced datasets. This research work focuses on an empirical comparative analysis of data augmentation techniques applied to time series analysis, in the domain of diabetic neuropathy detection. Building upon a preprocessed dataset, a suite of data augmentation techniques tailored to time series data are evaluated. Multilayer perceptrons and convolutional neural networks were trained using augmented datasets. Two strategies were employed for training and validation. Model performance was evaluated based on the ability to generalize from augmented to real-world data. These results suggest that data augmentation can be a feasible and reliable approach for prediction diabetic neuropathies based on postural time series data.
In software engineering, the effective reuse of components reduces costs, improves software quality, and accelerates development. This paper explores the lessons learned from a project to enable collaborative management of variability in software product lines (SPLs). This work aims to facilitate the collaborative management of variability, thus enhancing teamwork in software design. We implemented a software module in the FMweb and VariaMos tools to manage the variability of an SPL collaboratively. This module enables stakeholders to create variability models in real-time shared spaces collaboratively. The benefits of this approach include improved communication among team members, enhanced decision-making processes, and increased accuracy in managing variability. The findings illustrate the effective implementation of a collaborative module, enabling real-time, shared, and managed variability models. This advancement optimizes the development process, reduces costs, and represents a significant step forward in SPL engineering by providing a platform for improved collaborative work in software development.
A fuel cell is an electrochemical device that transforms chemical energy into electrical energy and heat. It occurs through a controlled chemical reaction. Among its essential components are the porous electrodes, which play a pivotal role in the operation of the fuel cells. These electrodes are crucial because they facilitate the efficient flow of reactants to the reaction sites, thereby driving the energy conversion process. The electrodes are made of porous materials characterized by intricate and varied microstructures. Given the complexity of these structures and the significant cost associated with experimental equipment, it is highly beneficial to perform computational studies as a preliminary step. This study focuses on reconstructing a reliable three-dimensional (3D) volume of porous electrodes from two-dimensional (2D) images. Such a reconstruction is essential for conducting predictive analyses of the electrodes' experimental performance and behavior. The study examines key microstructural parameters, including porosity, tortuosity, and diffusivity. The results of employing advanced computational techniques illustrate that the proposed methodology is highly effective and robust for generating accurate 3D reconstructions from 2D images. This capability offers significant advantages for understanding and optimizing the performance of fuel cell electrodes before embarking on costly experimental trials.
Understanding and trusting the model's predictions have become fundamental, especially in critical and scientific contexts. Stellar spectra classification is essential in astrophysics, especially for massive stars. This work focuses on deep-learning models for classifying Ha spectral lines of massive stars into 21 classes and the conformal prediction approach for analyzing uncertainty. We built a deep neural network classifier using the ISOSCELES database. We used 342,780 spectral lines, and the experimental results showed an overall accuracy score of 0.92 and an average F1 score of 0.94. We comprehensively described the conformal prediction approach used to obtain the prediction sets. We generated calibration datasets of different sizes, computed the respective conformal scores, and used different error rates to evaluate diverse confidence levels. The small prediction sets showed confident predictions and demonstrated the model's effectiveness in maintaining high classification accuracy while providing reliable measures of uncertainty.
Performance modeling of parallel applications is essential for optimizing resource usage in high-performance computing (HPC) systems. However, some scientific applications exhibit irregular performance behaviors, which complicates cre-ating accurate characteristic models. This irregularity is mainly due to these applications' nondeterministic computational and communication patterns. Tools such as PAS2P (Parallel Application Signatures for Performance Prediction) are used to extract detailed information about parallel applications. PAS2P is based on the repetitive behavior of the application to analyze and predict the application's performance, using the same resources that the parallel application uses for its execution. This paper presents a characterization model based on the PAS2P methodology for irregular applications that groups the repeatability patterns of all the processes running the application into a single characteristic model. To achieve this, we consolidate the different characterizations performed by each process independently, using metrics such as the number of instructions, the execution time of relevant sections, and the topological characteristics of the application. By grouping these repeatability patterns of all processes, we can obtain a concise and accurate representation of the behavior of irregular applications, thus improving predictability and performance optimization in HPC systems.
This study leverages evolutionary computation, particularly Particle Swarm Optimization (PSO) and its fuzzy variant (FST-PSO), to infer the weight matrix and threshold values required for threshold Boolean networks to reach a fixed-point attractor state, where all nodes converge to either 0 or 1. The research investigates the efficacy of these algorithms in generating networks that exhibit consensus properties, analyzing the topology and time steps needed to reach consensus. The results indicate that while PSO's effectiveness dropped significantly with increasing network size, achieving only 79% effectiveness for networks with eight nodes, FST-PSO maintained 100% effectiveness across all sizes. FST-PSO also demonstrated faster convergence, requiring fewer iterations and showing better scalability and stability in optimizing network parameters for consensus formation. This work contributes to understanding how network topology influences consensus formation, offering insights applicable to decision-making, optimization problems, and complex system analysis.
Designing academic curricula in universities is crucial for optimising student workload distribution. Excessive academic workload could negatively affect students' performance. The Balanced Academic Curriculum Problem (BACP) involves distributing the academic workload evenly across periods while prerequisites are satisfied according to an optimisation criterion. Traditionally, curriculum managers have manually balanced workloads, a time-consuming process often resulting in suboptimal distributions. Since different conflicting criteria have been proposed in the literature, this work considers a multi-objective approach to the BACP. Specifically, we designed and implemented three multi-objective metaheuristics: Non-dominated Sorting Genetic Algorithm II (NSGA-II), Multi-objective Simulated Annealing (MOSA) and Multi-objective Greedy Algorithm (MOGA), utilising datasets from literature and real-engineering cases in Chile. The results show the superior effectiveness of the NSGA-II algorithm, highlighting the potential of a resolution method to support curriculum design and optimisation in higher education. This approach enhances the student experience by creating manageable and balanced academic workloads.
The increase in thyroid cancer cases registered from 1990 to the last decade by 20 % worldwide, has created a need for faster and more accurate diagnostic tools for thyroid nodules. Deep Learning architectures, including Convolutional Neural Networks and Vision Transformers, have been developed to assist in diagnosing whether nodules are benign or malignant. Data Augmentation techniques such as DCGAN were also utilized. Hybrid models combining these type of architectures were also trained. The hybrid model developed by Google, re-trained with the TNCD dataset, achieved the best results with values of 77.20% for Accuracy, 77.97% for Recall, and 67.65% for Precision. The success of the hybrid approach depends on the architectures combined and whether they have been pre-trained. These findings suggest that a pre-trained model combining CNN and ViT is superior to using them independently, highlighting the potential of combining these architectures for improved the pre-diagnostic accuracy.
The climate at universities affects students' education and well-being. A comfortable, non-discriminatory educational environment is associated with greater well-being and higher persistence and graduation rates, especially in STEM fields where women tend to be minorities and face less inclusive environments. This study examines perceptions of university climate from a gender perspective in the Computer Engineering program. 288 of 509 first- and second-year students reported both perceptions of educational climate and factors of gender discrimination and emotional well-being. The results show that women perceive more negative factors than men, with a significant difference in ‘mental health problems’ (34% more for women). Women also report a higher incidence of sexist comments and inappropriate behaviors. In addition, one in three women have considered giving up their career, compared to one in four men.
