federated learning

Zero-Knowledge Governance: Cryptographic Compliance Auditing for Federated Edge AI

The rapid expansion of Federated Learning (FL) within Edge AI has enabled decentralized networks to collaboratively train sophisticated models without sharing raw, privacy-sensitive data [8]. However, this strict adherence to data privacy creates a significant "blind spot" for AI governance. Under emerging regulatory frameworks like the EU AI Act, regulators require proof of data safety, bias mitigation, and privacy preservation [4]. The current paradigm forces a paradox: regulators cannot verify whether individual edge nodes complied with these laws without violating the very privacy FL is designed to protect. In this paper, we propose a novel Zero-Knowledge Governance (ZKG) framework. By integrating Zero-Knowledge Succinct Non-Interactive Arguments of Knowledge (zk-SNARKs) into the local edge training phase, edge devices can cryptographically prove that their local datasets and model updates comply with encoded governance policies, without ever revealing the underlying data [1]. We present the architectural design of ZKG, analyze the vulnerabilities it addresses in decentralized AI, and propose optimizations to make cryptographic auditing viable for resource-constrained edge hardware.

Training Without Gathering the Data: A Critical Survey 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.

Decentralized Verification of Federated Learning Models

Federated Learning (FL) has emerged as a promising paradigm for training machine learning models on decentralized data sources without direct data sharing. However, ensuring the integrity and accuracy of these models remains a significant challenge. Traditional verification methods often rely on centralized aggregation, inherently compromising user privacy and introducing a single point of failure. This paper proposes a novel blockchain-based system for decentralized verification of FL models. The system leverages cryptographic proofs, specifically zero-knowledge Succinct Non-Interactive Argument of Knowledge (zk-SNARKs), allowing nodes to independently verify model updates without revealing the underlying data. This approach eliminates centralized aggregation, enhancing privacy and establishing a verifiable, distributed ledger of model updates. The core claim is that current decentralized verification methods are computationally expensive and rely on centralized aggregation, thus, this system offers a truly decentralized and privacy-preserving verification framework.

Federated learning for renewable energy systems: Algorithms, applications, and deployment challenges

zk-DEAP-artifact-1.0

Secure aggregation requires operating on private inputs while preserving confidentiality, verifying that each input satisfies constraints, and avoiding trusted parties. This scheme appears in federated learning, as well as Internet of Things, operational technology, blockchain, smart-grid, and voting deployments. Existing protocols achieve at most two of these requirements. We present zk-DEAP, a protocol combining lifted ElGamal homomorphic encryption, FROST threshold cryptography, and zero-knowledge proofs (Bulletproofs, zk-SNARKs, or zk-STARKs) for fully decentralized, verifiable, privacy-preserving aggregation.Our core technical contribution is a cryptographic binding mechanism preventing substitution attacks, where a valid proof for one input is paired with a ciphertext encrypting a different, malicious value. For Bulletproofs, we bind via ciphertext-derived randomness modification. For zk-SNARKs/zk-STARKs, to bind across algebraic structures, a well-known challenge in practical cryptography, we introduce a sophisticated blinded arithmetic hash commitment with an embedded Schnorr constraint, forcing the ZKP's private witness to equal the encrypted value, ensuring any circuit-verified constraint applies to the ciphertext.We implement three production-ready, defense-in-depth variants in Rust and evaluate across 5-500 participants on Azure VMs. Compared to unverified aggregation, zk-DEAP adds modest overhead while providing cryptographic guarantees against input manipulation that statistical defenses cannot match. Results show Bulletproofs, despite limited flexibility, achieve the fastest proof generation, while zk-STARKs provide post-quantum security with larger sizes, and zk-SNARKs achieve a middle ground.

Decentralized Federated Learning with Byzantine Fault Tolerance via Proof-of-Stake

Federated learning (FL) offers a promising approach to training machine learning models on decentralized data sources without directly exchanging the data itself. However, existing FL systems are susceptible to Byzantine attacks, where malicious participants intentionally submit incorrect or misleading updates, compromising the integrity and accuracy of the global model. This paper proposes a novel decentralized federated learning system that integrates Byzantine fault tolerance and a Proof-of-Stake (PoS) incentive mechanism to mitigate these vulnerabilities. The core idea is to incentivize participants to contribute honest updates while simultaneously providing cryptographic proofs to verify the validity of these updates. This system constructs a decentralized network where participants stake resources (e.g., computational power or storage) to gain voting rights and rewards for contributing correct updates. Malicious behavior is penalized through a reduction in stake and potentially exclusion from the network. The system utilizes cryptographic techniques, such as zero-knowledge proofs, to ensure the integrity of updates without revealing the underlying data. This approach significantly enhances the robustness of FL systems against adversarial attacks and promotes a more trustworthy and reliable learning environment. The proposed system addresses the critical challenge of Byzantine fault tolerance in FL, offering a practical and scalable solution for deploying FL models in sensitive environments.

FedVigil: Defending backdoor attacks in federated learning via dynamic Wasserstein-HDBSCAN vigilance

Differential Privacy for Federated Learning with Adaptive Noise

Federated learning (FL) presents a promising paradigm for training machine learning models on decentralized data sources without directly exchanging the data itself. However, the inherent privacy risks associated with aggregating model updates introduce significant challenges. Traditional differential privacy (DP) techniques often rely on adding uniform random noise to gradients, which can be overly conservative and degrade model accuracy. This paper introduces an adaptive noise scheme for FL that dynamically adjusts the noise level based on the sensitivity of the aggregated gradients. The proposed method monitors gradient variance and employs a stochastic gradient descent (SGD) variant with a dynamically adjusted learning rate and noise scale. We demonstrate through theoretical analysis and a simplified simulation that this approach significantly reduces the overall noise level compared to standard DP while maintaining a comparable privacy guarantee, ultimately leading to improved model accuracy in FL settings. The key contribution lies in the intelligent adaptation of noise, responding directly to the data's inherent characteristics.

Telemetry Dataset for Federated Learning on Heterogeneous Edge Devices Using Raspberry Pi 5 and NVIDIA Jetson Nano

This dataset accompanies the manuscript "Telemetry-aware federated learning on heterogeneous edge devices: an experimental study". The dataset contains telemetry measurements collected from a federated learning testbed consisting of a Raspberry Pi 5 and an NVIDIA Jetson Nano running the Flower federated learning framework. Telemetry was sampled at one-second intervals and includes CPU utilization, CPU frequency, CPU temperature, memory usage, disk activity, network activity, system load, process counts, and Jetson-specific GPU metrics where available. The experiments comprise four stages: baseline system stress profiling, local model training on MNIST, Fashion-MNIST, and CIFAR-10 datasets, federated learning experiments under varying local-epoch settings, and controlled hardware throttling experiments to investigate device heterogeneity. The dataset supports reproducibility of the statistical analyses presented in the associated manuscript and may be useful for research on federated learning, edge computing, telemetry analytics, resource-aware scheduling, heterogeneous computing, and performance characterization of embedded AI platforms. The data are provided in Microsoft Excel format with accompanying metadata and are intended for academic and research use.

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.

Telemetry Dataset for Federated Learning on Heterogeneous Edge Devices Using Raspberry Pi 5 and NVIDIA Jetson Nano

This dataset accompanies the manuscript "Telemetry-aware federated learning on heterogeneous edge devices: an experimental study". The dataset contains telemetry measurements collected from a federated learning testbed consisting of a Raspberry Pi 5 and an NVIDIA Jetson Nano running the Flower federated learning framework. Telemetry was sampled at one-second intervals and includes CPU utilization, CPU frequency, CPU temperature, memory usage, disk activity, network activity, system load, process counts, and Jetson-specific GPU metrics where available. The experiments comprise four stages: baseline system stress profiling, local model training on MNIST, Fashion-MNIST, and CIFAR-10 datasets, federated learning experiments under varying local-epoch settings, and controlled hardware throttling experiments to investigate device heterogeneity. The dataset supports reproducibility of the statistical analyses presented in the associated manuscript and may be useful for research on federated learning, edge computing, telemetry analytics, resource-aware scheduling, heterogeneous computing, and performance characterization of embedded AI platforms. The data are provided in Microsoft Excel format with accompanying metadata and are intended for academic and research use.

Replication package for Flows and Rounds: benchmark and retrieval scripts for distributed neurodynamic optimization and federated learning

This package contains the code and generated data behind the quantitative results of the survey "Flows and Rounds: A Survey of Distributed Neurodynamic Optimization and Federated Learning", submitted to Neurocomputing. It reproduces every number, table, and figure in Sections 2.3 and 8 of the article. No third-party, proprietary, or personal data is involved: all data is synthetic and generated at run time from fixed random seeds, so the package is self-contained and deterministic. Two scripts are included. The first, toy_benchmark.py, runs a controlled experiment on a synthetic distributed optimization problem with five nodes, twenty dimensions, and a closed-form optimum, averaged over ten seeds. It measures two things. Experiment one measures consensus drift as a function of the number of local steps per communication round, holding total local work fixed, and tests the prediction that drift grows with the dwell interval and vanishes when the agents hold identical objectives. Drift is found to be identically zero without gradient dissimilarity and to grow as E^0.88 and E^0.84 at the two heterogeneity levels tested. Experiment two measures whether an integral-enhanced flow, of the kind used in the zeroing neural network literature to reject persistent disturbance, also attenuates the Gaussian noise that a differential privacy mechanism deliberately injects. Both flows are driven by an identical noise stream and the noise response is isolated by differencing each noisy run against its own noise-free run. The integral term is found to remove steady-state bias almost entirely while leaving injected-noise energy unchanged, a ratio of 1.007. This second result contradicted the authors' expectation when the experiment was designed, and it is reported as measured. The second script, search_protocol.py, records a structured literature retrieval executed on 31 August 2026 and computes the deduplication and overlap counts reported in Section 2.3 of the article. It also emits a screening sheet listing all eighty-five retrieved records with their query provenance. The package is intended for readers who wish to verify the article's numerical claims, and for anyone extending the benchmark to other flows, noise models, or heterogeneity regimes. It requires only Python, NumPy, and Matplotlib.