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
