federated learning

Decentralized Machine Learning via Federated Learning with Byzantine Fault Tolerance

Federated learning (FL) offers a promising approach to distributed machine learning, enabling model training on decentralized devices without directly exchanging sensitive data. However, the inherent vulnerability of FL systems to Byzantine failures – where malicious participants intentionally corrupt the learning process – poses a significant threat to its reliability and security. This paper proposes a decentralized machine learning framework leveraging federated learning with Byzantine fault tolerance (BFT). The core claim is the development of a robust FL system capable of maintaining accurate and secure model training despite the presence of adversarial actors. The proposed mechanism utilizes verifiable computation and secret sharing techniques to mitigate the impact of Byzantine attacks. This work addresses a critical gap in existing FL research by explicitly tackling the security challenges associated with adversarial behavior, paving the way for more trustworthy and resilient distributed learning applications. The system is designed to ensure data privacy and model integrity, even when compromised by malicious nodes. The theoretical framework and the proposed architecture will be discussed in detail.

Performance Analysis of a Decentralized Federated Learning System for Spoken-Command Recognition: Resilience and Security Considerations

In the industrial edge-to-cloud continuum, data is often privacy-sensitive and spans multiple organizations that do not fully trust one another, making central aggregation of raw data undesirable and often non-compliant with regulations such as the General Data Protection Regulation (GDPR). Federated Learning (FL) addresses this by sharing model updates rather than raw data, but conventional FL assumes a central coordinator, leaving it exposed to poisoning and inference attacks and to a single point of trust and failure. Decentralized Federated Learning (DFL) couples FL with Distributed Ledger Technologies (DLTs), removing the coordinator and enabling verifiable aggregation in trustless, cross-organizational environments. In this work, we assess the applicability of DFL to on-device spoken-command recognition—a representative edge audio task underpinning voice-driven industrial interfaces—by comparing decentralized and centralized training under idealized and adversarial conditions. Using a Convolutional Neural Network (CNN) replicated across edge nodes, we evaluate resilience to inter-node data imbalance, to poisoning attacks, and to a privacy-preserving noise-injection defense against inference attacks, together with model compression for resource-constrained edge devices. The system pairs this comparison with a validation-based poisoning defense in which each node scores its peers’ updates on its own held-out data, and an update is aggregated only if a majority of nodes report a weighted F1-score above a threshold—requiring neither a shared validation set nor a trusted validator. Our results indicate that the DFL system achieves accuracy comparable to centralized baselines in most scenarios (weighted F1-score 0.762 across nine nodes, against 0.896 centralized), and that a cross-node validation mechanism reliably excludes poisoned updates as long as fewer than half of the nodes are compromised (within 3.54% of the unpoisoned model). Noise-based inference defenses reduce accuracy substantially (44.7% on average at a noise standard deviation of 1.0), exposing a sharp privacy–utility trade-off, whereas model compression preserves performance (0.765 against 0.762 for pruning and format conversion, with 8-bit quantization costing up to a further 13.3%). These findings clarify both the promise and the current limitations of decentralized, privacy-preserving learning for the industrial edge-to-cloud continuum.

Artificial intelligence in precise surgical planning for adolescent idiopathic scoliosis: applications and challenges

Adolescent idiopathic scoliosis (AIS) is the most common three-dimensional spinal deformity among adolescents, characterized by a coronal curvature of > 10°, often accompanied by vertebral rotation and abnormal sagittal alignment, which can lead to severe cardiopulmonary and neurological dysfunction. For moderate-to-severe or progressive AIS (e.g., Cobb angle ≥ 45°), surgical intervention is the primary treatment modality. However, traditional surgical planning relies heavily on surgeons’ clinical experience and is subject to substantial interobserver variability, making it difficult to fully meet the needs of personalized precision medicine. In recent years, advances in artificial intelligence (AI), particularly in machine learning and deep learning, have provided new methodological tools for precise surgical planning in AIS. This article reviews the main applications of AI in AIS surgical planning, including AI-assisted assessment and automated measurement of preoperative spinal deformities; personalized decision-making for surgical approaches and fusion levels; personalized biomechanical and finite element analysis (FEA)-based optimization of internal fixation devices; prediction of postoperative outcomes; and recent advances in 3D surgical simulation and navigation. Furthermore, this review examines the current limitations of translating AI models into clinical practice, such as data heterogeneity, the “black-box” nature of algorithms, and ethical and regulatory issues. It also discusses future directions, including multicenter federated learning and human-in-the-loop collaborative decision-making, and aims to provide a systematic overview to support intelligent and precise surgical treatment of AIS.

Artificial Intelligence in Smart Photovoltaic Systems for High-Efficiency Energy Conversion: A Systematic Review

Artificial intelligence (AI) is transforming photovoltaic (PV) systems from passive generators into self-forecasting, self-diagnosing, and self-optimizing energy assets. This review synthesizes 170 Scopus-indexed documents published between 2020 and 2026 across six technical domains: solar resource and PV power forecasting; intelligent monitoring, fault diagnosis, and cybersecurity; AI-driven control and design optimization; smart grid integration and real-time energy management; AI-assisted PV materials, devices, and manufacturing; and cross-sectoral smart PV applications. Quantitative synthesis shows that hybrid deep learning forecasters reduce root mean square error by 31.9–43.9% relative to persistence and single-model baselines. Attention-based architectures achieve mean absolute percentage errors of up to 5%. Machine learning classifiers achieve fault detection accuracies of 92.3–99.4% with protection response times below 100 ms, enabling predictive maintenance at the fleet scale. Reinforcement learning energy management lowers electricity costs by 20–55%, reduces peak demand by 13–31.5%, and raises PV self-sufficiency from 71.5% to 89.7%. AI-optimized thermal and material interventions deliver efficiency gains of up to 22.2%, and indoor perovskite devices exceed 40% conversion efficiency. Reported performance metrics are derived from heterogeneous datasets, horizons, and baselines; they are therefore synthesized as indicative ranges rather than directly comparable benchmarks. Persistent barriers include data scarcity, model opacity, cybersecurity vulnerabilities, edge deployment constraints, and energy justice concerns. These barriers are mapped to research directions in explainable AI, federated and transfer learning, digital twins, and blockchain-enabled energy markets. A roadmap to 2040 consolidates the findings and charts the transition toward high-efficiency, resilient, and sustainable solar energy conversion.

From data heterogeneity to convergence: A data-centric review of federated learning

Federated Learning (FL) has emerged as a promising solution for data hunger in centralized learning. This paradigm enables privacy with multiple clients to train a shared-task model collaboratively without exposing their local data. While being a key component in any learning system, data is also a primary source of vulnerabilities and challenges, and a major determinant of a stable and well-converged training. Existing FL reviews describe general foundations, security practices, opportunities, challenges, and applications, without delving into diverse aspects of data and considering problems from the data perspective. They rarely provide a data-lens synthesis that links concrete data properties, split protocols, and defenses to convergence speed and stability. This survey fills that gap with three advances. First, we analyze “non-IID” into measurable traits and rank their influence on convergence as strong, medium, or light, explaining the mechanisms behind each and reconciling evidence across images, texts, and graphs. Second, we connect experimental splitting practices to the real phenomena they emulate, expose the artifacts they introduce, and show how those artifacts affect target accuracy. Third, we analyze how data-related vulnerabilities and their proposed defenses affect convergence, reporting performance under clean and adversarial conditions to make the convergence-robustness trade-off explicit. To our knowledge, this is the first survey to provide a complete understanding of data-related challenges that govern FL. With clear takeaways distilled for each concern, our work serves as actionable guidance, helping practitioners design their system with predictable convergence and stability.

Federated learning: From design foundations to advanced participant selection paradigms

Federated Learning (FL) has emerged as a promising paradigm owing to its significant advantages over conventional centralized machine learning approaches. It is a decentralized learning framework in which a global model is iteratively constructed by aggregating locally trained models from distributed clients, thereby ensuring enhanced data privacy while reducing communication overhead and latency. This survey provides a comprehensive yet concise overview of the state-of-the-art in FL covering key architectural aspects, i.e., scale of federation, data distribution, network topology, exchanged information, training procedures, machine learning models, and aggregation techniques. It further reviews widely used FL frameworks and datasets for simulation and evaluation, incorporating an analysis of dataset selection strategies, the limitations of synthetic data partitioning, and the gap between simulated and real-world deployments. In addition, the incorporation of FL in various applications for their rapid acceleration is discussed. The challenges pertinent to FL are examined under two main categories, i.e., general challenges and those specific to the participant selection process. Building on this, the survey positions participant selection as a central design component in FL systems and provides a unified analysis of state-of-the-art methods through the lenses of composability, trade-offs, and practical design considerations, highlighting its critical role in achieving efficient, robust, fair, and scalable FL.

A Survey of What to Share in Federated Learning: Perspectives on Model Utility, Privacy Leakage, and Communication Efficiency

Federated learning (FL) has emerged as a secure paradigm for collaborative training among clients. Without data centralization, FL allows clients to share local information in a privacy-preserving manner. This approach has gained considerable attention, promoting numerous surveys to summarize the related works. However, the majority of these surveys concentrate on FL methods that share model parameters during the training process, while overlooking the possibility of sharing local information in other forms. In this paper, we present a systematic survey from a new perspective of what to share in FL, with an emphasis on the model utility, privacy leakage, and communication efficiency. First, we present a new taxonomy of FL methods in terms of three sharing methods, which respectively share model, synthetic data, and knowledge. Second, we analyze the vulnerability of different sharing methods to privacy attacks and review the defense mechanisms. Third, we conduct extensive experiments to compare the learning performance and communication overhead of various sharing methods in FL. Besides, we assess the potential privacy leakage through model inversion and membership inference attacks, while comparing the effectiveness of various defense approaches. Finally, we identify future research directions and conclude the survey.

Federated learning for multi-institutional AI in healthcare via digital pathology

Abstract The integration of artificial intelligence (AI) in digital pathology has shown significant promise in advancing cancer diagnostics, grading, and treatment response prediction. However, widespread development and deployment of robust AI models face critical challenges due to data silos, privacy concerns, and the need for large-scale multi-institutional datasets. Federated Learning (FL) presents a transformative approach by enabling collaborative model training across hospitals without direct data sharing. In this review, we summarize recent developments in FL as applied to digital pathology, highlight pioneering use cases, and explore the technical, regulatory, and ethical hurdles. We discuss how FL can enable scalable, privacy-preserving AI models, and outline future directions for standardizing and validating FL-based approaches in clinical workflows.

Smart aquaponics: trends, challenges, and future directions

Abstract Smart aquaponics couples recirculating aquaculture with hydroponic plant cultivation, and a substantial body of recent research applies IoT sensing, machine learning, and edge computing to this domain. The literature, however, lacks a synthesis that maps how predictive models, system architectures, and biological context combine in deployed systems, leaving researchers and practitioners without a clear technical roadmap. This systematic literature review (SLR) addresses this gap through 10 research questions following the Kitchenham and Charters protocol and PRISMA-style screening. The search initially identified 3,123 records from six bibliographic databases and snowballing; after duplicate removal, screening, and quality assessment, 49 primary studies were retained for analysis. First, the literature has a prediction-to-control gap: 24% of the studies report forecasters or classifiers without specifying how the resulting prediction is consumed by an actuator, leaving inference layers technically ahead of control layers. Second, system architectures are transitioning from centralised cloud designs toward federated and edge configurations to address privacy, latency, bandwidth, and sensor-drift issues. Third, hybrid models incorporating physical, or domain knowledge show promising performance under non-stationary aquaponics conditions. Fourth, long-term and multi-site field deployments remain very limited, making it difficult to interpret reported machine-learning accuracy as evidence of commercial readiness. Overall, future progress requires stronger integration of prediction, control, calibration, biological benchmarking, and operator usability.

Six-Hour Sepsis Prediction in Intensive Care: Patient-Level Nested Validation of Structured, Centralized Multimodal, and Simulated Federated Models

Federated Quantum-Inspired Tensor Compression for Energy-Efficient Edge Learning in Bandwidth-Constrained IoT Swarms

A privacy-preserving federated spatiotemporal dynamic graph neural network framework for epileptic seizure prediction

Abstract Epilepsy ranks among the most prevalent and debilitating neurological disorders, globally affecting an estimated 50 million individuals across all age groups and socioeconomic backgrounds [WHO, 2019]. In recent years, research on its prediction methods has made significant progress, driven by advancements in artificial intelligence technology. Epilepsy prediction models based on electroencephalogram (EEG) signals and deep learning have become an important research direction in the field of neuroscience, and related research results have shown an exponential growth trend. However, there are still several key problems that need to be solved in existing research: First, the mainstream model architecture are focused on traditional neural network framework, exclusively trained in centralized settings that are incompatible with the privacy regulations and data-sharing constraints governing real-world clinical environments; second, the deep learning approaches, including recent Graph Neural Network (GNN) models, using static graph modeling methods, which ignore the dynamic network topological evolution characteristics of EEG signals in the time-varying process, fails to effectively explore the high-order nonlinear correlation characteristics contained in the topological structure of brain functional networks; To address this limitation, this paper proposes a patient-dependent privacy-preserving Federated Learning framework that integrates an epilepsy prediction model DygonNet based on spatiotemporal dynamic graph neural network, a local learning model at each federated client, deployed within a cloud-based simulation environment. In the proposed architecture, the CHB-MIT, SWEC-ETHZ and the TJU-HH iEEG datasets are treated as three independent federated clients representing distinct clinical sites, each performing federated training on their private EEG data, while a central cloud server simulated on Google Colab Pro aggregates the model updates using the Federated Averaging (FedAvg) and FedProx algorithms without ever accessing raw patient recordings. The model defines the dynamic graph structure of EEG signals, innovatively introduces the Transformer model and dynamic graph neural network into the field of epilepsy prediction to fully learn the temporal and spatial characteristics of EEG signals, and proposes a hierarchical graph pooling mechanism based on the attention mechanism in a Federated environment. Experiments show that the model shows excellent epilepsy prediction performance on both public and private datasets.