Javier Prieto

Background Federated Learning (FL) enables collaborative model training across distributed clients without sharing raw data, making it attractive for privacy-sensitive energy forecasting. However, standard aggregation algorithms, most notably FedAvg and FedProx, assume stationary client data distributions, an assumption routinely violated in multi-national power grids where seasonal patterns, demand crises, and policy changes induce persistent regime shifts. Methods We present RAFA (Regime-Aware Federated Aggregation), a novel FL framework that explicitly detects and adapts to distribution shifts in client model updates without accessing raw client data. RAFA comprises three components. These are a Mahalanobis-distance detector operating on random-projected update vectors; reliability-weighted aggregation that exponentially discounts shifted clients while preserving exact FedAvg behaviour when no shift is detected; and always-on personalised head fine-tuning that adapts each client’s final prediction layer to its local regime. Results We validate RAFA on a new multi-national benchmark of six years (2019–2024) of hourly electrical load data for eight European bidding zones sourced from the ENTSO-E Transparency Platform. RAFA achieves a test MAE of 185.4 MW, an 11.8% improvement over FedAvg (210.2 MW), with by far the largest gain for the client exhibiting the strongest seasonal contrast (France, − 30.5 % ) and consistent improvements across the remaining clients (e.g., Portugal − 7.6 % , Poland − 7.4 % ). Ablation studies confirm both mechanisms contribute independently. Conclusions RAFA provides a lightweight, privacy-preserving extension of standard federated aggregation that is robust to regime shifts in energy systems. The approach generalises to any federated setting with temporally non-stationary client distributions and a separable model architecture.

Los modelos epidemiológicos clásicos SEIR ofrecen interpretabilidad, pero presentan una flexibilidad limitada para capturar factores contextuales dinámicos, mientras que los enfoques de aprendizaje automático suelen carecer de transparencia. Presentamos SEIR-XAI, un marco híbrido que integra un modelo SEIR en tiempo discreto con aprendizaje automático explicable para inferir la tasa de transmisión efectiva β como una variable dependiente del contexto. Mediante modelos interpretables y explicaciones basadas en SHAP, SEIR-XAI permite un análisis tanto global como local de cómo las intervenciones políticas y los factores ambientales influyen en la dinámica de transmisión. La evaluación con datos de COVID-19 muestra una mejora en el ajuste del modelo en comparación con un modelo SEIR base, manteniendo al mismo tiempo la interpretabilidad. Los resultados ponen de relieve que una modelización epidémica eficaz no debe centrarse únicamente en la precisión predictiva, sino también en proporcionar información transparente y accionable para el apoyo a la toma de decisiones.

The usage of blockchain technologies incorporates a trust layer due to its immutability and transparency for all participants. The implementation of this technology in order to validate and authenticate the identity of different entities in a Data Spaces ecosystem, allows to all the contributors to validate which of them have certain sets of traits suitable for their data sharing, interests, and moreover, role managing. This way, connectors can automatically identify with who they are interacting and choose which data to share with them based on these characteristics. Some useful examples are the detection of certain issued verifiable credentials that prove their belonging to a specific sector, accreditations … that automatically provide them with access to specific datasets. In order to achieve this, the presented Trust Layer uses a governance infrastructure with blockchain, an identity layer to store private keys and connect the blockchain address to a decentralized identifier, and a credential manager to issue and verify the Verifiable Credentials. Added to this, an integration layer is designed and implemented, which connects these components to the Data Space Connector. For this specific technical implementation, the design has been built to be operable with Eclipse Dataspace Components (EDC), given its Verifiable Credentials interoperability and its wide adoption among Data Spaces as their preferred connector. A qualitative discussion of the architecture’s strengths and challenges is provided, covering aspects such as decentralization, regulatory alignment, and adoption barriers, together with a comparison against alternative identity and trust models for federated data spaces.

Background Intrusion detection at Internet of Things (IoT) edge gateways must run under tight compute and memory budgets, motivating dual-path designs that classify most flows with a cheap model and escalate only uncertain flows to a heavier one. A prior conference study introduced such a system, in which a tiny multilayer perceptron (MLP) on features compressed by principal component analysis (PCA) is the fast path, gated to a heavy MLP by the maximum softmax probability (MSP). That system recovers near-heavy accuracy while escalating few flows. Methods We extend that work into a systematic study of what signal should drive the escalation gate. On three benchmarks (ACI-IoT-2023, TabularIoTAttacks-2024, and the large-scale CICIoT2023), we compare routing signals (MSP, predictive entropy, margin, energy, temperature-scaled confidence, latent norm) against a learned router and an oracle bound at matched cost; study temperature-scaling calibration and a leakage-free threshold rule; test a tiny-ensemble disagreement signal; and evaluate robustness, latency, model size, statistical significance, and cross-dataset transfer. Results Confidence-gated routing gives large, statistically significant gains over the tiny model on every dataset, whereas a learned router improves only marginally over plain MSP, so once an escalation budget is fixed a simple MSP gate suffices. On CICIoT2023 we observe an oracle gap. Routing cannot reach the heavy model even at large budgets, because the tiny model is confidently wrong on a few high-volume attack families such as denial-of-service (DoS) floods that are never escalated. Neither ensembling nor confidence-independent signals close the gap. A fixed threshold does not transfer across datasets, its escalation rate swinging from 1 % to 55 % . Conclusions The efficiency benefit grows with a stronger heavy backbone, including a gradient-boosted-tree heavy. Gates should be set by a budget, and closing the oracle gap requires identifying confident errors within high-confidence classes rather than a better routing signal.

The usage of blockchain technologies incorporates a trust layer due to its immutability and transparency for all participants. The implementation of this technology in order to validate and authenticate the identity of different entities in a Data Spaces ecosystem, allows to all the contributors to validate which of them have certain sets of traits suitable for their data sharing, interests, and moreover, role managing. This way, connectors can automatically identify with who they are interacting and choose which data to share with them based on these characteristics. Some useful examples are the detection of certain issued verifiable credentials that prove their belonging to a specific sector, accreditations … that automatically provide them with access to specific datasets. In order to achieve this, the presented Trust Layer uses a governance infrastructure with blockchain, an identity layer to store private keys and connect the blockchain address to a decentralized identifier, and a credential manager to issue and verify the Verifiable Credentials. Added to this, an integration layer is designed and implemented, which connects these components to the Data Space Connector. For this specific technical implementation, the design has been built to be operable with Eclipse Dataspace Components (EDC), given its Verifiable Credentials interoperability and its wide adoption among Data Spaces as their preferred connector. A qualitative discussion of the architecture’s strengths and challenges is provided, covering aspects such as decentralization, regulatory alignment, and adoption barriers, together with a comparison against alternative identity and trust models for federated data spaces.

Background Federated Learning (FL) enables collaborative model training across distributed clients without sharing raw data, making it attractive for privacy-sensitive energy forecasting. However, standard aggregation algorithms, most notably FedAvg and FedProx, assume stationary client data distributions, an assumption routinely violated in multi-national power grids where seasonal patterns, demand crises, and policy changes induce persistent regime shifts. Methods We present RAFA (Regime-Aware Federated Aggregation), a novel FL framework that explicitly detects and adapts to distribution shifts in client model updates without accessing raw client data. RAFA comprises three components. These are a Mahalanobis-distance detector operating on random-projected update vectors; reliability-weighted aggregation that exponentially discounts shifted clients while preserving exact FedAvg behaviour when no shift is detected; and always-on personalised head fine-tuning that adapts each client’s final prediction layer to its local regime. Results We validate RAFA on a new multi-national benchmark of six years (2019–2024) of hourly electrical load data for eight European bidding zones sourced from the ENTSO-E Transparency Platform. RAFA achieves a test MAE of 185.4 MW, an 11.8% improvement over FedAvg (210.2 MW), with by far the largest gain for the client exhibiting the strongest seasonal contrast (France, − 30.5 % ) and consistent improvements across the remaining clients (e.g., Portugal − 7.6 % , Poland − 7.4 % ). Ablation studies confirm both mechanisms contribute independently. Conclusions RAFA provides a lightweight, privacy-preserving extension of standard federated aggregation that is robust to regime shifts in energy systems. The approach generalises to any federated setting with temporally non-stationary client distributions and a separable model architecture.

Underwater cultural heritage (UCH) is one of the richest and most vulnerable records of humankind’s past

Background Accurate wind power forecasting is essential for ensuring the stability and efficiency of modern energy infrastructures. However, the nonlinear behavior and complex correlations present in wind turbine data make prediction increasingly challenging. Recent advances in Quantum Machine Learning (QML) suggest that quantum kernels can capture higher-order feature interactions, potentially improving forecasting performance over classical models. Methods We propose a hybrid High-Performance Computing–Edge (HPC–Edge) forecasting framework that integrates quantum-enhanced Support Vector Regression (QSVR) executed on HPC resources with lightweight Linear Regression (LR) deployed at the edge, close to the data source. The system follows a residual-learning strategy in which the quantum model predicts and corrects the residual errors of the edge-based LR model. The approach is evaluated using a real-world wind turbine SCADA dataset containing more than 50,000 ten-minute interval measurements collected in Türkiye. Results The results show that while edge-based LR models remain competitive for local, low-latency forecasting, the hybrid HPC–Edge architecture substantially improves predictive performance. Across multiple sampling configurations, the hybrid model achieves an RMSE of 117.95, an R 2 of 0.9638, and a SMAPE of 27.00%, outperforming both standalone edge models and standalone quantum models. Conclusions These findings demonstrate that combining edge-level efficiency with quantum-enabled residual correction provides a practical and effective pathway for integrating quantum regression techniques into operational wind power forecasting systems. The hybrid HPC–Edge approach improves accuracy without compromising deployability, highlighting its potential for real-world renewable energy applications.