federated learning

基于自适应参数调整的联邦学习 (Adaptive Parameter Adjustment Federated Learning)

Federated Learning (FL) is a machine learning technique that enables training models on decentralized data sources while preserving data privacy. However, traditional FL approaches often rely on global model updates, which can be vulnerable to adversarial attacks and suboptimal performance on heterogeneous data. This paper proposes a novel framework called Adaptive Parameter Adjustment Federated Learning (APAFL) that dynamically adjusts model parameters based on local data characteristics, enhancing model robustness and privacy. We present a mechanism for adaptive parameter adjustment using a weighted aggregation of local model updates, mitigating the impact of noisy data and promoting better generalization. The proposed method demonstrates significant improvements in model accuracy and privacy compared to existing FL techniques.

Blockchain-Based Federated Learning with Differential Privacy

Federated learning (FL) offers a promising approach to training machine learning models on decentralized data sources without direct data sharing. However, traditional FL systems remain vulnerable to privacy breaches and data manipulation. This paper proposes a novel blockchain-based architecture that addresses these concerns by integrating data provenance tracking, integrity verification, and differential privacy mechanisms. The system leverages blockchain technology to create an immutable record of model updates and data contributions, ensuring transparency and accountability. Simultaneously, differential privacy techniques are applied during the training process to protect the privacy of individual data contributors. This combined approach significantly enhances the security and trustworthiness of FL systems, enabling secure and collaborative model training across diverse data sources. The proposed system utilizes cryptographic hashing and Merkle trees to guarantee data integrity and employs noise injection strategies within differential privacy mechanisms to protect user data. The core contribution is a secure and verifiable FL framework with enhanced privacy guarantees.

Hybrid AI-Blockchain Frameworks for Secure Spectrum Data Governance in Smart Grids

The emerging 6G technologies will bring billions of IoT sensors which are going to demand little radio spectrum. The resulting high level of connectivity poses some novel problems, among them network congestion, and privacy: because energy-consumption data might be shared unintentionally, it provides insight into personal behavioral patterns. In addition, the disproportionate access to highly developed infrastructure can contribute to the need to continue the digital divide between urban and rural areas. This paper presents a decentralized infrastructure combining federated learning and Hyperledger Fabric to deal with these issues. To implement it, PyTorch is used to perform distributed learning tasks, and MATLAB is used to generate the synthetic spectrum traces that reflect the real-world CBRS conditions based on the NTIA field measurements between 2021 and 2024. These traces were validated by Kolmogorov–Smirnov test ( p = 0.87). The framework outperforms conventional centralized DSS benchmarks while complying with 3GPP Release 18 transparency and ETSI ESG principles. The framework is a modular, open-source system with sharding that scales horizontally to over 5,000 nodes. The proposed solution successfully addresses the scarcity of the spectrum with the use of cooperative resource sharing through congested urban vehicle-to-everything (V2X) networks, remote solar-powered microgrids, and so on. The findings verify that the next generation 6G networks are capable of not only providing extremely high data rates but also improved privacy and fair connectivity in different environments.

Federated Learning security with FedEDAuth: Embedding-Distribution Authentication for Counterfeit IC Detection

A Device-Agnostic Framework for Personalized Anomaly Detection in Wearable and Mobile Ecosystems: Applications in Personal Safety and Respiratory Health Surveillance

Abstract—This paper presents a reformulated, device-agnostic architectural framework for two interrelated applications: (i) personalized audio-based personal safety threat detection, and (ii) respiratory pattern anomaly detection for early-stage epidemiological screening. Originating from an ideation dialogue, the proposed concepts are critically re-examined against fundamental constraints in machine learning, embedded systems, privacy engineering, and sensor heterogeneity. We propose a generalized abstraction layer that decouples sensing modality from inference logic, enabling deployment across heterogeneous smart devices (wearables, smartphones, IoT nodes) without hardware-specific dependencies. We address the zero-positive-example training problem through a reformulation as one-class classification and temporal anomaly detection. Privacy is preserved via on-device federated feature extraction with no raw data transmission. We further analyze the practical limitations identified by domain experts and propose mitigations grounded in current literature. The framework is positioned as a foundational architecture rather than a deployable product, inviting interdisciplinary validation. Impact Statement—By introducing a sensing abstraction layer and reformulating personal safety and respiratory monitoring as one-class anomaly detection problems, this work provides a portable, privacy-preserving architecture that can operate across heterogeneous consumer devices without hardware-specific redesign. The framework addresses the zero-positive-example constraint inherent to rare-event detection and offers a structured research agenda for empirical validation, potentially accelerating the development of reliable, edge-deployed health and safety monitoring systems. Index Terms—Anomaly detection, one-class classification, device-agnostic computing, respiratory pattern analysis, personal safety systems, federated learning, wearable computing, edge inference.

**Decentralized Federated Learning with Byzantine Fault Tolerance using Graph Protocols**

This record contains a research paper preprint and its associated PDF document.

Federated Learning with Differential Privacy: Enhancing Privacy in Decentralized Model Training

Federated learning (FL) presents a promising paradigm for training machine learning models on decentralized data sources without directly exchanging data. However, the inherent collaborative nature of FL still poses privacy risks. This paper investigates the integration of differential privacy (DP) into the FL framework to provide an enhanced level of privacy protection. We introduce a novel approach where noise is injected into model parameters during the training process, adhering to the principles of DP. This mechanism mitigates the risk of information leakage from individual user data while maintaining model accuracy. We formally define the proposed algorithm and analyze its privacy-utility trade-off. The results demonstrate that the incorporation of DP significantly strengthens the privacy guarantees of FL, offering a robust solution for privacy-sensitive applications. Our work contributes to the growing body of research focused on secure and privacy-preserving FL techniques.

Adaptive federated edge intelligence with semantic communication and trust-aware optimization for heterogeneous IoT networks

The rapid growth of heterogeneous Internet of Things (IoT) networks introduces significant challenges in communication efficiency, energy consumption, and reliable distributed intelligence due to diverse device capabilities, dynamic channel conditions, and limited computational resources. Existing bit-level communication and conventional federated learning approaches often suffer from excessive communication overhead, inefficient resource utilization, and unstable model convergence under non-IID data and unreliable device participation. To address these limitations, this paper proposes FEI-SemCom, an adaptive federated edge intelligence framework that integrates semantic communication, federated learning, and joint resource optimization for heterogeneous IoT systems. The proposed framework introduces a heterogeneity-aware semantic encoder, adaptive semantic compression mechanism, semantic contribution-based client selection strategy, and trust-aware aggregation scheme to improve communication efficiency and learning robustness. Unlike existing approaches that optimize communication, learning, and resource allocation independently, FEI-SemCom jointly adapts semantic representation, client participation, transmission parameters, and aggregation weights according to device capability, channel quality, and energy availability. Furthermore, the proposed framework improves resilience under intermittent connectivity conditions by enabling adaptive client participation, reliability-aware aggregation, and resource-aware communication decisions, allowing federated edge intelligence to operate effectively despite temporary link failures and unstable device availability. Extensive simulation results demonstrate that the proposed framework achieves 91.2% accuracy at 5 dB signal-to-noise ratio (SNR), converges within approximately 120 training rounds, and reduces communication overhead, energy consumption, and latency to 45 MB, 48 J, and 110 ms, respectively. Compared with conventional approaches, FEI-SemCom reduces communication overhead by up to 38% while maintaining stable convergence and robust learning performance. These results demonstrate the potential of adaptive semantic communication combined with federated edge intelligence for scalable and resource-efficient IoT applications.

Federated Medical Image Segmentation under Real-World Label Noise: A Benchmark Suite for Noisy Label Learning Method Selection

Abstract While federated learning (FL) enables collaborative medical image segmentation without centralizing sensitive data, real-world deployment is frequently complicated by cross-site label imperfections such as contour disagreement, missing or additional structures, and confused labels. Federated noisy label learning (FNLL) aims to mitigate these effects, yet remains underused in practice as existing evidence is largely based on synthetic noise, simplified settings, and limited real-world noisy evaluation. We address this gap by introducing a benchmark suite that combines diverse real-world noisy datasets, deployment-relevant client-noise scenarios, and label-noise-targeted evaluation to support systematic FNLL assessment and informed method selection. The suite combines curated real-world noisy medical image segmentation datasets from diverse sources with a comprehensive federated segmentation framework including various client-noise scenarios and noise-targeted evaluation. To demonstrate its capabilities, we compare representative FNLL methods across approaches, including noise-aware aggregation, robust personalization, label correction, and sample selection. In-depth data analysis shows that real-world segmentation label noise occurs both in isolation and in combination of characterized noise types. The benchmark shows that FedSelect performs strongest within noise-targeted evaluation, underlines FedAvg as a competitive baseline, and provides a practically informed decision guide for FNLL method selection based on label-noise type, client-noise scenario, and technical-cost considerations. The presented suite provides a realistic and discriminative basis for FNLL evaluation in medical image segmentation and establishes a reusable foundation for fair benchmarking, dataset-specific label-noise characterization, and future method development under realistic federated settings. Code is available at https://github.com/MIC-DKFZ/FedSegNoiseBench .

Training Without Gathering the Data: A Historical Development Review of Federated Learning and Privacy-Preserving AI

This article presents a narrative review of Federated Learning and Privacy-Preserving AI in the context of Artificial Intelligence. The literature on this topic has expanded substantially over recent decades, yet it remains fragmented across subfields, methods, and national research traditions. Drawing on an interpretive synthesis of representative contributions, the review reconstructs the historical development of the area, examines the conceptual foundations and definitional disputes that organize its debates, and maps the contemporary landscape of research, including the methodological shift toward data-intensive approaches and the institutional pressures that shape publication practice. Particular attention is given to the role of federated learning and privacy as organizing themes, and to the conditions under which findings from different research traditions can be brought into productive comparison. The review identifies three synthetic conclusions: the literature is cumulatively strong but organizationally weak; methodological pluralism is better understood as a resource than as a defect; and the growing practical salience of the topic raises the stakes of its unresolved conceptual questions. An agenda for future work is proposed, emphasizing integrative research designs, transparent synthesis practices, and the protection of definitional and infrastructural work on which cumulative progress depends. The article is intended as both a reference map for newcomers and a provocation for specialists in Artificial Intelligence.

Training Without Gathering the Data: A Historical Development Review of Federated Learning and Privacy-Preserving AI

This article presents a narrative review of Federated Learning and Privacy-Preserving AI in the context of Artificial Intelligence. The literature on this topic has expanded substantially over recent decades, yet it remains fragmented across subfields, methods, and national research traditions. Drawing on an interpretive synthesis of representative contributions, the review reconstructs the historical development of the area, examines the conceptual foundations and definitional disputes that organize its debates, and maps the contemporary landscape of research, including the methodological shift toward data-intensive approaches and the institutional pressures that shape publication practice. Particular attention is given to the role of federated learning and privacy as organizing themes, and to the conditions under which findings from different research traditions can be brought into productive comparison. The review identifies three synthetic conclusions: the literature is cumulatively strong but organizationally weak; methodological pluralism is better understood as a resource than as a defect; and the growing practical salience of the topic raises the stakes of its unresolved conceptual questions. An agenda for future work is proposed, emphasizing integrative research designs, transparent synthesis practices, and the protection of definitional and infrastructural work on which cumulative progress depends. The article is intended as both a reference map for newcomers and a provocation for specialists in Artificial Intelligence.

A Privacy-Preserving Distributed Machine Learning Aggregation Scheme

Federated Learning enables data to remain on local devices, yet malicious servers can still infer sensitive user information by analyzing client-uploaded model updates, posing significant privacy leakage risks. Existing secure aggregation schemes—such as Differential Privacy, Homomorphic Encryption, and traditional Secret Sharing—struggle to achieve an ideal balance among model accuracy, computational/communication overhead, and adaptability to complex aggregation scenarios. To address this, this paper proposes a multi-server secure aggregation scheme based on additive secret sharing. The scheme introduces multiple non-colluding servers; each client splits its local model update into random secret shares and distributes them to these servers. Each server independently performs weighted aggregation, and the global model is reconstructed by a leader server. Theoretical analysis demonstrates that, owing to the linear homomorphism of additive secret sharing, the aggregation result of our scheme is mathematically equivalent to the standard FedAvg algorithm. Experimental results on the MNIST dataset show that our scheme achieves a test accuracy of 86.13% after 30 global rounds, closely matching the baseline FedAvg (86.12%), with only a 6% additional time overhead per round. Under reasonable non-collusion assumptions, the scheme achieves information-theoretic privacy protection, effectively breaking the trilemma among privacy, accuracy, and efficiency.Keywords:Federated Learning, Additive Secret Sharing,Secure Aggregation; Privacy Preservationords.