Spatial planning is key to guiding the revitalization of rural territories like “España Vacía” towards sustainable development. However, current tools are insufficient to address the complexity of the Doughnut Economics framework. This research, following a Design Science Research (DSRM) approach, proposes a methodology and a geo-technological workflow prototype. This system integrates Artificial Intelligence (AI), adapted Explainable AI (XAI), and interactive visualization to support local technicians. The objective is to operationalize the Doughnut, enabling systemic analysis and informed decision-making while prioritizing usability and free/open-source software. We hypothesize that the integration of adapted XAI and interactive visualization will significantly enhance trust, understanding, and adoption of AI tools by non-expert users. The expected contribution is a validated, user-centred method connecting sustainability theory and advanced AI with local practice.
This paper introduces an innovative MLOps platform specifically designed for edge AI, addressing critical challenges posed by limited computational resources constraints at the edge. The platform targets environments where inference is performed at constrained edge devices, while training is executed at resourceful on-premise or cloud servers. Utilizing a dual pipeline architecture, the platform continuously enhances model accuracy through server labeling and retraining processes, automating updates while ensuring stringent deployment quality standards. This platform provides a scalable blueprint for deploying robust edge AI solutions in dynamic environments, effectively leveraging powerful cloud resources to significantly enhance edge AI model performance. Our framework integrates state-of-the-art tools, including Apache Airflow, MLflow, and Apache TVM, to provide seamless and efficient lifecycle management of AI models optimized for edge environments. The platform has been validated using a satellite-based sea ice classification test bed. Targeting satellite AI applications, we present a comprehensive platform that integrates automated pipelines for data ingestion, ground-based image labeling, model retraining, version control, and optimized deployment. The test bed demonstrates the platform’s effectiveness in improving model accuracy through iterative labeling and retraining processes. Validation involved retraining a baseline model using an expanded dataset that included previously unrepresented ice classes. The updated model demonstrates a marked improvement in accuracy and detailed classification compared to the baseline, highlighting the platform’s capability to iteratively improve models based on new incoming data.
This article presents a Ph.D. research proposal for the automation of Digital Twin construction in industrial contexts through the semantic integration of heterogeneous data. The approach combines Large Language Model with the Asset Administration Shell framework to extract and map technical information from structured and unstructured sources (such as sensors, manuals and ERP/MES systems) into standardized submodels. The methodology includes four stages: data collection, semantic mapping using, organization into submodels and integration into Digital Twins. Initial tests with simulated data show the ability of LLMs to identify equivalent technical terms and generate structured data compatible with Asset Administration Shell. Ongoing work includes future activities with data from industrial partners, development of evaluation metrics and analysis with domain experts. The aim is to reduce manual modeling work, support interoperability and enable the construction of scalable Digital Twin in line with Industry 4.0 frameworks.
Human-Centered Machine Learning (HCML) models often face challenges due to inherent biases related to population variability and limited access to large datasets. This results in algorithms that fail to generalize and accommodate out-of-distribution samples, thereby hindering real-world applications. Additionally, standard training procedures tend to make neural networks vulnerable to privacy risks such as reconstruction attacks. To address these issues, we propose a novel training method based on an adversarial network that aims to reduce the representation bias induced by the lack of diversity among training samples. Unlike similar approaches that use a known bias predictor as the adversarial signal, our method mitigates multiple unknown biases, acting as an effective regularization term that reduces the validation gap while also removing non-essential features. This feature selection further improves privacy by preventing the model from being repurposed or used to retrieve information about training or inferred samples, as demonstrated on the IMDb-Face dataset where the method achieves approximately a 6.7
Artificial intelligence (AI) is rapidly transforming the financial sector, but there is a notable lack of research on the integration of gender perspectives in AI financial applications. This systematic literature review, conducted using the PRISMA methodology, examines existing studies and identifies three primary areas of focus: innovation, regulation, and decision-making. Despite the insights provided by these areas, the review reveals a significant gap in addressing gender considerations in the development and deployment of AI technologies in finance. The study highlights the need for further research to incorporate gender perspectives into these key areas, ensuring the creation of more inclusive and sustainable AI financial applications. Future work should aim to bridge this gap, fostering AI advancements that are equitable and beneficial for all members of society.
The increasing use of digital tools in education has opened new opportunities to enhance student engagement, particularly for neurodivergent learners who face challenges with traditional instructional methods. Attention measurement is a critical aspect of adaptive learning, yet current approaches rely on static models that fail to capture real-time cognitive fluctuations. This paper presents a literature review on AI-driven attention tracking and adaptive learning systems. We analyse key methodologies, including eye-tracking, head-pose estimation, and machine learning-based engagement models, highlighting their advantages and limitations. Our findings emphasize the need for more robust, personalized AI systems that dynamically adjust content to improve student concentration. This review sets the foundation for future research on AI-driven adaptive education tailored to individual cognitive profiles.
The quality of data is of paramount importance when training a Machine Learning model, especially in sensitive domains. Since model performance is directly dependent on data quality, it is critical to ensure that high-quality data samples are used during training. In Federated Learning, however, the inability to assess data directly limits our understanding of its quality, which can undermine confidence in the model’s reliability. This issue is of particular importance in domains, such as those dealing with medical records, where data integrity is essential. To address this challenge, we propose a novel approach to quality assurance in Federated Learning, called FedQAM (Federated Quality Assurance Mechanism). Our method evaluates the performance of client models on a publicly available test set and compares their results against a predefined threshold at the client side before incorporating them into the global model. This ensures that only reliable client updates are aggregated. Additionally, we introduce and evaluate various threshold selection schemes to optimise the process. Our approach effectively identifies and filters out unreliable or malicious clients, preventing them from compromising the model’s integrity.
The rapid proliferation of AI-powered edge devices necessitates robust benchmarking frameworks to assess performance, fairness, and reliability across diverse environments. dAIEdge-VLab, a collaborative platform for launching benchmarking experiments on remote AI-enabled edge devices, requires a trustworthy and decentralized approach to ensure integrity, transparency, and security. Blockchain technology presents a compelling solution by enabling unchangeable benchmarking experiment reports, decentralized consensus mechanisms, and tamper-proof data management. Through smart contracts, automated execution of benchmarking protocols can enhance reproducibility and fairness while reducing dependence on centralized authorities. Additionally, token-based incentive models can foster active participation and resource-sharing among stakeholders. Despite these advantages, integrating blockchain into remote AI benchmarking poses challenges, including increased computational overhead, scalability constraints, and latency issues. Moreover, the cost of on-chain storage and transaction fees can impact feasibility, requiring hybrid solutions that combine off-chain storage with blockchain validation. This paper explores the opportunities and trade-offs of blockchain integration in dAIEdge-VLab, presenting a novel decentralized framework that balances security, efficiency, and accessibility for AI benchmarking on remote edge nodes. By leveraging blockchain’s strengths while addressing its limitations, we propose an architecture that fosters trust, transparency, and collaboration in AI performance evaluation on edge devices.
This study presents a neural network-based model specifically designed to optimize the sizing of photovoltaic (PV) systems for industrial applications, focusing on maximizing self-consumption and enhancing return on investment. Given the growing role of solar PV in the energy transition, effective sizing of installations is essential to balance energy generation, economic return, and environmental sustainability. By leveraging local solar irradiation data, monthly demand profiles, and specific PV configurations, the model predicts the optimal PV capacity across diverse commercial and industrial scenarios in various regions of Spain. Scenario-based testing demonstrates the model’s accuracy in aligning PV output with site-specific demand, contributing to reduced grid dependency and operational costs. This adaptable approach offers a practical solution for businesses aiming for energy autonomy and eco nomic efficiency through tailored PV system designs, underscoring the role of neural networks in advancing energy management strategies.
Music generation is a complex and challenging problem that has made significant progress through recent machine learning solutions. The challenge lies not only in rendering natural-sounding audio but also in capturing the underlying musical structure. This paper presents a real-time system that generates a musical accompaniment for an input lead melody in the MIDI format. In this paper, we introduce REMIBlock, a novel tokenization approach, and show that it is suited for accompaniment generation. Our method uses the GPT-2 architecture [11], optimized for efficient on-device performance, to generate accompaniments that are rhythmically and harmonically coherent with a musician’s performance. The model used in this papers reaches a perplexity of 7.69 on the test set, proving the model’s ability to understand musical language. Additionally, using two metrics—groove and scale consistency—proposed in [3], we show that the generated accompaniments closely match the ground truth, differing by at most 3.4 https://github.com/UncleBen420/JazzyGPT2 .
Classroom behavior analysis is a key component of multimodal learning analytics, which has advanced alongside the digitalization of education, artificial intelligence, and hardware. Several studies have proposed systems to enhance learning outcomes in both online and face-to-face environments. However, hardware solutions to address critical deployment issues remain limited, particularly during face-to-face lessons. These issues include high bandwidth consumption, latency, computing demand, data security, privacy, and anonymity. This work presents a hybrid edge computing model for analyzing student behavior and indoor environmental variables during in-person lessons using AI-powered systems. Our model addresses network, computing resources, and data-related challenges. Tested in three case studies with different classroom and lesson configurations, the results demonstrate the model’s capability to process HD video streams of over 20 students per frame with four deep neural networks at five frames per second, reducing bandwidth consumption from 0.85 Mbps to 0.1 Kbps, conserving computing resources, and ensuring data privacy. Additionally, the model supports indoor environmental variable acquisition, real-time dashboards, and human-computer interaction on the same node device.
Managing the life cycle of electric motorcycles requires efficient monitoring and maintenance strategies. This paper presents a mobile tool to streamline and optimize this process, providing access to key vehicle information, structured maintenance logs, and an organized record of component usage. Unlike traditional methods, which often rely on fragmented or manual record-keeping, our solution integrates key functionalities into a single, user-friendly platform. The prototype leverages mobile technology to enhance accessibility, ensuring that manufacturers, dealerships and repair shops can make data-driven decisions. By improving operational efficiency and reducing maintenance errors, this tool contributes to the sustainability and longevity of electric motorcycles. This paper details the development process and key features demonstrating its potential to transform electric vehicle lifecycle management.
Updating a neural network running on an edge device can be quite challenging, especially when the device uses a low bandwidth uplink communication channel. For that reason, sending all the model weights becomes prohibitive. An example of such a device is an Earth Observation (EO) satellite, which employs Convolutional Neural Networks (CNNs) to make decisions related to the images captured using its EO sensors. A possible approach to efficiently update these kinds of devices is to leverage the fact that the weights which are currently used by the edge device are known and update only a subset of them. In this work, several criteria for choosing the subset of weights which are updated have been compared. Since our main focus is to efficiently update models running on EO satellites, the datasets which have been used to test the methods correspond to images taken by these kinds of satellites. The results show that the best approach is Deep Partial Update (DPU). Specifically, the fully fine-tuned model achieves a Dice score of 0.6641 but if DPU is applied in a setting where 25
This article discusses the proposed study for the completion of a Ph.D. thesis. A system to remotely monitor patient progress in rehabilitation using IoT devices and mobile platforms is proposed. This system feeds advanced artificial intelligence (AI) algorithms and techniques such as federated learning. Additionally, real-time posture estimation with computer vision is implemented. This set of technologies enables accurate monitoring of rehabilitation exercises without the need for constant supervision. This approach is intended to benefit both the healthcare professional, by reducing the burden associated with routine check-ups, and the patient, who can be assured of continuous monitoring of their injury and a rapid response to any abnormalities in recovery.
Traffic safety is a significant issue, particularly for motorcyclists, who are significantly more vulnerable to fatal accidents compared to drivers of enclosed vehicles. Advanced Rider Assistance Systems have the potential to improve motorcyclist safety by mitigating critical risks like rear-end collisions. This paper proposes an integrated Advanced Rider Assistance Systems architecture using the RTDETR object detection model to predict rear-end collisions and provide real-time alerts to the riders. Specifically, it aims to mitigate critical risks like rear-end collisions, offering a comprehensive safety solution. The experimental data indicate that the proposed system is suitable for diverse operational environments, supporting the development of advanced sensing and alert systems to improve motorcyclist safety. Our system achieves an average precision (AP) of 0.688 at IoU=0.50 and reduces collision risk by issuing timely warnings. Contributions include the integration of RTDETR for improved detection accuracy, a multi-threshold warning mechanism, and a detailed analysis of system performance under varied conditions.
The problem I intend to address in my research is the lack of ability of conventional Intrusion Detection Systems to properly handle the complexities and changing nature of Industrial Internet of Things environments. Existing IDS systems, which are primarily based on deep learning approaches, usually face high computing costs, limited real-time adaptability, and a lack of transparency in decision-making processes. Is it possible to create an adaptable and interpretable system that uses several data sources and hybrid decision models to detect intrusion in IIoT networks better? In this study, I describe a new multilayered intelligent decision support system that uses data fusion, fuzzy logic, and hybrid decision models. This approach is meant to improve detection accuracy while eliminating false situations.
This article presents a real-time object detection and distance estimation system implemented on a low-cost platform. The system uses a Raspberry Pi 5 and two cameras in a stereoscopic configuration to capture pairs of images. Object detection is performed using YOLO neural networks and distance estimation is based on the disparity between the centers of the detected bounding boxes. The system is evaluated in terms of detection performance, inference speed and depth estimation accuracy. Three YOLO models (YOLOv8n, YOLO11n and YOLO12n) are tested at different resolutions. Among them, the YOLO11n with a resolution of 320 × 320 achieves the best balance between processing speed and detection quality in stereoscopic operation. The system has a low error in depth estimation at close range, with absolute errors of less than 1.2 cm up to 60 cm. At greater distances, accuracy is affected by the reduction in the size of the bounding box, which limits the reliability of the disparity. Possible improvements include using segmentation-based localization and optimizing the stereo configuration. The proposed system is suitable for short-range applications in controlled environments and serves as a basis for future improvements in embedded vision systems.
The design and development of consumer products require an interdisciplinary approach, often constrained by time-consuming prototyping and manual decision-making processes. As product complexity increases and market demands evolve, the need for automation and intelligent collaboration becomes evident. This paper presents a case study on the design and virtual validation of a premium pen using a multi-agent system, leveraging the integration of large language models (LLMs) and software agents. This combination enables a rational representation of human expertise and interactions, streamlining the design process while enhancing adaptability. Using CrewAI, agents were configured with specialized tasks, collaborating to optimize design, select sustainable materials, and establish quality standards. The agents generated a markdown report and a 3D simulation using Blender and Python, ensuring efficient coordination for an ergonomic, sustainable, high-quality pen. By modeling the rational behavior of human experts, the system demonstrated how LLMs and multi-agent coordination can reduce decision overhead and improve collaboration. The results show that multi-agent systems streamline product development by reducing decision overhead, improving task delegation, and enhancing collaboration. The final design met strict virtual quality standards and aligned with market preferences. This study demonstrates the role of multi-agent systems and LLM integration in Industry 4.0, supporting digital prototyping and virtual simulations to replace traditional physical prototyping.
Urban mobility increasingly prioritizes sustainable and user-friendly transportation solutions, driving the need for advancements in vehicle design that enhance both efficiency and rider comfort. This paper presents the methodological framework with empirical validation of the A-Mover project, which aims to develop a high-autonomy electric motorcycle optimized for comfortable, efficient, and eco-friendly urban mobility. As a part of a multidisciplinary network, this study focuses on research related to vehicle ergonomics. To assess ergonomic conditions, a single data collection was conducted with experience motorcycle in urban environments. The study integrates biomechanical sensors and artificial intelligence software to evaluate rider posture and emotional responses. Data collection includes: (a) kinematic measurements from high-precision accelerometers, (b) physiological indicators such as heart rate, (c) ergonomic analysis of both the motorcycle and rider, (d) subjective assessments of comfort and posture, and (e) facial emotion recognition. The collected data will offer preliminary insights to support future recommendations for improving rider comfort and safety, health, and well-being through multidisciplinary analysis and innovative technologies.
