Wael Issa, Nour Moustafa, Benjamin Turnbull, KimâKwang Raymond Choo
Sixth-generation (6G) wireless networks are set to transform the Internet of Things (IoT) by enabling faster, smarter, and more connected systems. These networks will bring together a wide range of devices, including cars, robots, industrial machines, and smartphones, to support edge intelligence and real-time decision-making. Federated learning (FL) supports this shift by allowing devices to collaboratively train models without sharing raw data, which helps to protect user privacy. Despite its advantages, FL faces significant security challenges, including poisoning attacks and Byzantine clients, both of which can compromise the training process and degrade the accuracy and reliability of the global model. Although existing methods can detect malicious updates, many advanced attacks still bypass statistical defenses relying on metrics such as median and distance. Thus, developing an FL system that ensures both reliable decision-making and privacy and security guarantees in IoT networks remains a significant challenge. This study introduces a Digital Twin-driven Blockchain-enabled Federated Learning (DT-BFL) framework designed for IoT networks. The framework creates a digital representation of the IoT environment to support secure and decentralized edge intelligence using blockchain and federated learning technologies. DT-BFL is built to detect and filter out potentially poisoned model updates from malicious participants. This is achieved through a new smart contract-enabled decentralized aggregation method called Local Updates Purify (LUP). LUP uses a two-stage filtering process: First, it applies Median Absolute Deviation (MAD) to initially remove outliers, then uses statistical features and clustering to separate honest from malicious updates before aggregating the global model. It also assigns a Trust Score (TS) to each participant based on how much their updates differ from the global model and then uses a genuine criterion to select honest clients by evaluating trust scores, update similarity, and deviation from the global model. Experimental results show that DT-BFL effectively defends against various poisoning attacks on datasets like MNIST, ToN-IoT, and CIFAR-10 using models such as CNN, MLP, ResNet, and DenseNet, and maintains high accuracy even when 50% of the clients are malicious. Using a permissioned blockchain further secures the system by enabling aggregation of the decentralized model and authentication of clients through smart contracts. The source code is available on https://github.com/UNSW-Canberra-2023/LUP .
Mohammad Kamrul Hasan, S. Rayhan Kabir, Shayla Islam, Salwani Abdullah · 8 authors
The Smart Grid Artificial Intelligence-of-Things (SG-AIoT) uses SCADA (Supervisory Control and Data Acquisition) and AMI (Advanced Metering Infrastructure) to improve electricity distribution. However, the current SG-AIoT system faces challenges such as inaccurate energy forecasting, insecure data computation, information leakage, and data CIA (Confidentiality, Integrity, and Availability) issues. Smart-meter data has been provided to drive initiatives such as the UKâs "Low-Carbon-London" project, where the need for accurate forecasting and grid data security is felt. To address these issues, this paper proposes a secure federated learning framework that combines LSTM (Long Short-Term Memory) neural networks with AES (Advanced Encryption Standard) cryptographic blockchain across three-tier grid layers (Edge smart meters, Fog SCADA servers, and Cloud server). An AES-based blockchain framework is developed to secure data communication across all grid layersâedge, fog, and cloud. At the IoT smart meter edge nodes, a double-layered LSTM and dropout neural network is used for accurate energy demand forecasting. Forecasted data from each edge node is encrypted and aggregated at the fog layer using the AES blockchain. The proposed CBFedAggSum (Cryptographic Blockchain with Federated Learning Aggregation through Summation) method enables secure and efficient model aggregation at fog and cloud nodes. This end-to-end process strengthens grid data security and improves forecasting accuracy, effectively addressing key challenges in SG-AIoT systems. We compared our approach with six existing methods (FedAvg, FedSGD, SecAgg, 1D-CNN-GRU, PP-CE-FL, and Edge-Cloud-AIoT-FL). We found that our model outperformed them in forecasting accuracy and privacy preservation.
The rise of Industry 5.0 focuses on merging advanced intelligence, automation, and human-centered teamwork in industrial settings. However, keeping interconnected IoT networks secure is still a challenging problem. This paper proposes a new security framework that combines Blockchain, Federated Transfer Learning, and zero trust network (ZTN) principles to improve IoT security in Industry 5.0. Blockchain is a decentralized ledger that ensures secure data sharing and protects model updates. Federated Transfer Learning allows model training across distributed IoT devices to keep data private. The ZTN approach enforces strict access rules, assuming that no entity is trusted by default. The proposed framework offers a scalable and resilient solution to protect next-generation industrial IoT networks, using Blockchain for data security, transfer learning for adaptability, and ZTN for strict access control. The ZTN architecture strengthens security by checking every access request and keeping the IoT system safe. The experimental results show good performance of the proposed method, with better accuracy, precision, recall, and F1 scores. The model achieved an accuracy of 0.85, 0.88, and 0.87 for learning rates of 0.01, 0.001, and 0.0001, respectively, at 100 epochs. The precision values reached 0.84, 0.87, and 0.86, while the recall scores were 0.82, 0.86, and 0.85, respectively. The F1-scores were recorded at 0.83, 0.86, and 0.85, which confirms the robustness of our model.
Federated Learning is a transformative approach to building secure and resilient AI systems by enabling decentralized model training without exposing raw data. As part of Challenges and Solutions for Cybersecurity and Adversarial Machine Learning, this chapter examines its role in enhancing cybersecurity and mitigating adversarial threats, emphasizing its privacy-preserving capabilities and robustness against attacks. Key security challenges, including adversarial model poisoning, communication risks, and data privacy concerns, are analyzed alongside solutions such as differential privacy, secure aggregation, and robust optimization techniques. The discussion extends to Federated Learning's applications in critical sectors such as healthcare, finance, and edge computing, where secure AI deployment is essential. Addressing these challenges and proposing viable solutions, the chapter provides a comprehensive perspective on Federated Learning's potential to enhance AI security and resilience in adversarial environments.
Ahmed Sayed M. Metwally, Yazeed Alhumaidan, Saad Alzahrani, Mohamed H. Abdelati
Implementing artificial intelligence (AI) and blockchain technology in management systems transforms traditional libraries into advanced information centers that are data-driven and effectively managed. While these technologies enhance efficiency and operational capabilities, they also present two critical challenges: data privacy and ethical concerns. This study examines the role of AI and blockchain in library management, focusing on issues related to data privacy and ethical challenges that arise from their use. It also offers best practices to ensure safe implementation. The research adopts a comprehensive mixed-methods approach, involving qualitative interviews and quantitative surveys to identify these challenges within the system architecture, assess the effectiveness of current designs, and propose a complete framework using privacy-preserving technologies. This framework incorporates innovative cryptographic techniques, including homomorphic encryption, differential privacy, and zero-knowledge proofs, providing a novel model for the ethical use of AI in libraries. The findings indicate that robust data protection, transparency, and accountability are essential to building trust in AI-powered library services.
Legacy centralized Electronic Document and Records Management Systems (EDRMS) are confronted with security vulnerabilities, single failure points, and regulatory compliance. This paper envisions a decentralized EDRMS with Secure Multi-Party Computation (SMPC) to ensure privacy-preserving data computation and a Federated Byzantine Agreement (FBA) for fault-tolerant consensus. Secure Multi-Party Computation provides secure computation with secret inputs without sacrificing individual data inputs, while the Federated Byzantine Agreement enhances system reliability by consensus despite faulty or malicious nodes. The proposed solution significantly increases data security, scalability, and compliance with regulatory requirements through cryptographic techniques and distributed protocols. The performance assessment demonstrates a significant rise in accuracy (93%), efficiency (92%), recall (94%), and F1-score (91%), performing better than traditional centralized models. The framework is a solid, privacy-compliant solution for businesses handling sensitive data, like healthcare and finance, to provide safe, resilient, and efficient document management in decentralized settings.
The works of Garg et al. [S&Pâ24] (aka hinTS) and Das et al. [CCSâ23] introduced the notion of silent threshold signatures (STS) - where a set of signers silently perform local computation to generate a public verification key. To sign a message, any set of t signers sign the message non-interactively and these are aggregated into a constant-sized signature. This paradigm avoids performing expensive Distributed Key Generation procedure for each set of signers while keeping the public verification key constant-sized.In this work, we propose the notion of committee-based silent threshold signature (c-STS) scheme. In a c-STS scheme, a set of signers initially perform a one-time setup to generate the verification key, and then a subset of signers are randomly chosen for an epoch to perform the threshold signing while the other signers are not authorized to sign during that epoch. This captures existing systems like Ethereum Altair and Dfinity where only a specific committee is authorized to sign in a designated epoch. The existing STS schemes cannot be extended to the committee setting because the signature verification only attests to the number of signing parties, not which committee they belong to.So, we upgrade hinTS to the committee setting by proposing Dyna-hinTS. It is the first c-STS scheme and it requires a onetime silent setup and generates a one-time public verification key that does not vary with the committee. Assuming a set of 1024 signers (with corrupt 682 signers), hinTS generates an aggregated signature in 1.7s whereas Dyna-hinTS generates it in 0.35s within a committee of 80 signers. This yields a 4.9Ă improvement over hinTS for signature generation at the cost of increasing signature verification time by 4% over hinTS. Dyna-hinTS supports general access structure, weighted signatures and improves existing multiverse threshold signatures.
As Internet technology develops swiftly, the significance of privacy protection is escalating in the realms of e-commerce, e-government and software security. Due to the combination of the benefits of proxy signatures and blind signatures, the proxy blind signature scheme not only distributes the workload across application networks but also effectively safeguards the confidentiality of sensitive information. Additionally, the identity-based proxy blind signature protocol can avoid the problem of heavy certificate management and is widely used in electronic commerce and other scenarios. However, some identity-based proxy blind signature protocols that rely on the large integer factorization problem and the discrete logarithm problem are unable to withstand from attacks from quantum computers. Furthermore, current lattice-based proxy blind signature protocols offer only heuristic security and require three rounds of information exchange during the signing phase. In this paper, we introduce a new two-round identity-based proxy blind signature scheme based on lattices. This scheme utilizes a zero-knowledge proof protocol on lattices as its core component to develop an interactive two-round signature scheme that is free from security proof vulnerabilities. Additionally, we validate the security of the proposed protocol within the random oracle model and conduct a performance analysis.
Karthikeyan Marappan, C L Annapoorani, Nazura Javed, Udhaya Sankar T P · 6 authors
Enhancing digital security has become essential in the age of decentralized systems. Zero-Knowledge Proof (ZKP) is a cryptographic method that allows one party to demonstrate ownership of certain knowledge without disclosing the actual data. When used with blockchain, Zero-Knowledge Proofs significantly improve privacy, validation, and data secrecy. This study investigates the use of Zero-Knowledge Proof in blockchain environments to enhance trustless authentication, secure transactions, and identity safeguarding. The main aim is to provide safe conversations while preserving transparency and anonymity. The immutable characteristics of blockchain, together with the privacy-preserving protocols of zero-knowledge proofs, allow applications like secret smart contracts, decentralized identification systems, and safe voting methods. The collaboration between blockchain and zero-knowledge proofs mitigates data exposure concerns and diminishes dependence on conventional centralized verification. This strategy aids in the development of robust digital infrastructures that can withstand cyber-attacks and data breaches. The incorporation of Zero-Knowledge Proof in blockchain signifies a progressive improvement in the protection of sensitive digital transactions across several industries.
Zhaowei Hu, Jin Rui-fang, HangYi Quan, Shiyun Ni · 5 authors
To address privacy leakage risks arising from low collaborative user engagement, third-party trust deficits, and insufficient collaboration timeliness in location-based services (LBS), this paper proposes a dual-protection framework integrating blockchain technology and threshold cryptography for safeguarding location privacy. The framework employs asymmetric encryption with Shamir's (t, n) secret sharing to encrypt user queries, distributing decryption key fragments to collaborative users while generating n anonymous service requests through location generalization strategies. A temporary private blockchain constructed using smart contracts ensures confidential data transmission, supported by a dynamic privacy parameter configuration system based on Byzantine fault tolerance. The framework implements a priority-response consensus mechanism through Token-based equity proof-of-stake, prioritizing service for users with higher Token values. To mitigate privacy breaches caused by unresponsive collaborators, a competitive incentive mechanism ensures timely information submission. Through ciphertext fragment verification algorithms and Lagrange interpolation-based key reconstruction, the framework enables secure query decryption and service matching in untrusted third-party environments, guaranteeing information security, integrity, and non-repudiation. Experimental validation using real-world datasets confirms the framework's feasibility and operational effectiveness.
Ensuring privacy and security in healthcare data sharing is critical due to the sensitive nature of patient information and the growing threat of cyber attacks. This paper explores the development of privacy-preserving data-sharing protocols for healthcare systems by integrating cryptographic techniques and blockchain technology. The study aims to establish a secure framework that facilitates seamless data exchange among healthcare stakeholders while maintaining data integrity, confidentiality, and access control. Key cryptographic mechanisms, including homomorphic encryption, zero-knowledge proofs, and attribute-based encryption, are employed to ensure that only authorized entities can access patient records without exposing sensitive details. Blockchain technology is leveraged to create a decentralized and tamper-resistant ledger, ensuring transparency and auditability in data-sharing transactions. Smart contracts are utilized to enforce predefined access policies automatically, enhancing security and compliance with regulations such as the General Data Protection Regulation (GDPR) and the Health Insurance Portability and Accountability Act (HIPAA). The findings indicate that the proposed framework significantly mitigates risks associated with unauthorized access, data breaches, and single points of failure. Comparative analysis with traditional centralized systems demonstrates improved efficiency, scalability, and security in healthcare data management. The integration of blockchain and cryptographic techniques ensures robust privacy-preserving mechanisms without compromising accessibility or interoperability. This research provides a novel approach to secure data sharing in healthcare, fostering trust among stakeholders while ensuring compliance with privacy regulations. Future work will focus on optimizing computational efficiency and addressing scalability challenges to facilitate widespread adoption in real-world healthcare ecosystems.
Daria Schumm, Cedric von Rauscher, Katharina Olga Emilia MĂŒller, Burkhard Stiller
Transparency and immutability of blockchains can expose metadata and raise concerns about its classification as personal data under privacy regulations. This paper investigates privacy risks associated with metadata in blockchain-based identity systems. Additionally, two privacy-preserving mechanism designs, namely Zero-Knowledge Proof (ZKP) and Homomorphic Encryption (HE), to protect metadata are proposed. As a result, this work introduces the first use case of HE privacy-preserving mechanism in the context of Decentralized Identity (DI) and Self-Sovereign Identity (SSI) systems.
The digital economy is one of the most dynamic, fastest growing, and impactful domains in modern economic development. To fully exploit the potential value of data as an economic resource, many governments have implemented policies to encourage and regulate data circulation and trade. Despite the increasing adoption of blockchain-based data trading platforms, a significant challenge persists: verifying data quality without compromising privacy. This deficiency reduces data utilization efficiency and limits its economic potential We propose HES, a decentralized scheme for data quality verification in transactions, which integrates homomorphic encryption, blockchain, and zero-knowledge proofs to balance privacy and verifiability. The proposed method introduces data meta-certificates to represent the original data and employs cryptographic methods to enable zero-knowledge verification. This mechanism supports the verification of both data content and quality while preserving privacy. Data buyers can verify datasets without accessing specific content or involving third-party intermediaries. In this decentralized framework, data buyers can independently verify whether a dataset meets their requirements based solely on its content and quality indicators. Our experiments show that the proposed approach solves the challenge of reaching data quality consensus in distributed scenarios, enabling data buyers and sellers to swiftly align on quality standards. This facilitates the secure circulation and trade of data assets while unlocking their economic value. The findings offer a robust, privacy-preserving consensus mechanism for evaluating data quality in transactions.
The Artificial Intelligence of Things (AIoT) is driving human society from âinformationâ to âintelligenceâ, and the information technology industry is undergoing tremendous changes. However, AIoT data faces security threats such as leakage and illegal access when assisted by third parties. Therefore, some scholars use attribute-based proxy re-encryption (ABPRE) for secure sharing of data. However, the existing ABPRE schemes suffer from high computational overhead and inefficient attribution revocation, which seriously hinders practical application. To solve these problems, in this paper, we propose an efficient attribute-based proxy re-encryption scheme with fast attribute revocation (EPREAR). We design a non-interactive zero-knowledge proof protocol based on blockchain to ensure the verifiability of the key during attribute revocation. Furthermore, we devise a boundless encryption and decryption mechanism to enable the system's encryption and decryption with a fixed computation overhead, regardless of the size of the attribute set. And EPREAR possesses the ability to add infinite attributes without re-initializing the system. Finally, we perform theoretical and experimental analyses that show EPREAR has excellent computational performance. As a consequence, it has better application value in AIoT.
Management of secret keys for digital signatures is one of the most critical issues in decentralized applications. Since there is no administrator, losing a secret key can result in losing all assets or rights. To address this problem, fuzzy extractors and fuzzy signatures, which generate private keys directly from biometric information, have been considered in addition to conventional biometric authentication. However, these methods using biometric secret keys do not support group signatures. Therefore, it is not applicable to use cases that require consensus building by a specific community (group), such as DAO and DeFi.In this paper, we propose a new scheme for converting existing fuzzy signatures to group signatures using zero-knowledge proofs to address this problem. More precisely, we first define an anonymizable signature that is a generalization of a group signature and then convert a fuzzy signature into an anonymizable signature using an ordinary (classical) zero-knowledge proof. In addition, the signature data size is optimized to a constant size using zk-SNARK. Our implementation experiments show that our schemes achieve practical signature generation and verification times and signature sizes even for a group of up to 100,000 people. This paperâs results can be used to prevent the loss of secret keys and enable flexible DApps use cases.
As global energy markets undergo digitization and decentralization, multinational finance consortia increasingly rely on artificial intelligence (AI) to analyze geospatial and sensor data for investment modeling, risk assessment, and infrastructure optimization.However, cross-border data exchange poses significant privacy, security, and sovereignty concerns-particularly in energy-sensitive contexts where geospatial telemetry and environmental sensor networks contain critical operational intelligence.This paper advances a secure federated learning (FL) architecture tailored for multinational energy finance consortia, leveraging encrypted, AI-driven analytics to harmonize data utility and confidentiality.The proposed architecture integrates homomorphic encryption, differential privacy, and secure multi-party computation within a federated learning framework.It enables collaborative AI model training across sovereign entities and private stakeholders without transferring raw data, thus preserving jurisdictional control while enabling unified forecasting of energy supply, climate impact, and financial risk metrics.The paper details a tiered security model that accommodates variable data sensitivity levels-from satellite imagery and wind turbine telemetry to emission sensors and power grid diagnostics.Furthermore, the study presents a pipeline that processes heterogeneous datasets-such as LIDAR scans, thermal signatures, and remote sensor logs-using encrypted deep learning models capable of geospatial segmentation, anomaly detection, and predictive trend inference.Emphasis is placed on maintaining model accuracy in non-IID (non-independent and identically distributed) data scenarios, a common feature in distributed energy infrastructure.Policy implications are explored through case studies involving regional green bond issuance, multinational solar grid financing, and climate-resilience investments.The paper concludes with a governance blueprint for secure AI collaboration in energy finance ecosystems, balancing transparency, performance, and regulatory compliance.
Federated learning (FL) is increasingly recognised for addressing security and privacy concerns in traditional cloud-centric machine learning (ML), particularly within personalised health monitoring such as wearable devices. By enabling global model training through localised policies, FL allows resource-constrained wearables to operate independently. However, conventional first-order FL approaches face several challenges in personalised model training due to the heterogeneous non-independent and identically distributed (non-iid) data by each individual's unique physiology and usage patterns. Recently, second-order FL approaches maintain the stability and consistency of non-iid datasets while improving personalised model training. This study proposes and develops a verifiable and auditable optimised second-order FL framework BFEL (blockchain enhanced federated edge learning) based on optimised FedCurv for personalised healthcare systems. FedCurv incorporates information about the importance of each parameter to each client's task (through fisher information matrix) which helps to preserve client-specific knowledge and reduce model drift during aggregation. Moreover, it minimizes communication rounds required to achieve a target precision convergence for each client device while effectively managing personalised training on non-iid and heterogeneous data. The incorporation of ethereum-based model aggregation ensures trust, verifiability, and auditability while public key encryption enhances privacy and security. Experimental results of federated CNNs and MLPs utilizing mnist, cifar-10, and PathMnist demonstrate framework's high efficiency, scalability, suitability for edge deployment on wearables, and significant reduction in communication cost.
The rapid adoption of Federated Learning (FL) in privacy-sensitive domains such as healthcare, IoT, and smart cities underscores its potential to enable collaborative machine learning without compromising data ownership. However, conventional FL frameworks face several critical challenges: high computational overhead on edge devices, significant communication latency due to frequent model updates, vulnerability to model and data poisoning attacks, and limited privacy-preserving mechanisms that expose systems to inference risks. These issues hinder the scalability, efficiency, and trustworthiness of FL in real-world, large-scale deployments-particularly in domains like Electronic Health Records (EHR) management, where data sensitivity is paramount. To address these challenges, this paper introduces the Enhanced Privacy-Preserving Blockchain-Enabled Federated Learning (EPP-BCFL) framework, which integrates blockchain with hybrid privacy mechanisms and intelligent aggregation strategies. The architecture comprises three layers: (1) an Edge Nodes Layer for on-device learning; (2) a Federated Aggregation Layer using Secure Multi-Party Computation (SMPC) and Differential Privacy (DP); and (3) a Blockchain Layer with a lightweight PoS + BFT consensus mechanism. Experimental evaluation on CIFAR-10 demonstrates 95.2% accuracy, a 43% reduction in communication latency, a 37% decrease in computational cost, and robust defense against data/model poisoning and adversarial attacks. Attack resilience improved accuracy from 72.5 to 93.2%, while privacy budget tuning achieved 90.3% accuracy at Δ = 1.0. Compared to state-of-the-art models, EPP-BCFL exhibits superior performance in terms of security, scalability, and support for edge device heterogeneity, validating its applicability in secure EHR management.
Autonomous vehicles have attracted considerable attention from researchers and organizations, with artificial intelligence (AI) playing a key role in this technology. For AI models in autonomous vehicles to be reliable, the integrity of the training data is crucial, resulting in the development of various blockchain-based management systems. However, conventional blockchain systems incur significant time delays when processing training data transactions, posing challenges in autonomous vehicle environments that require real-time processing. In this study, we propose a hashgraph-based training data management system for trusted AI. To validate our system, we conducted simulations using the CARLA simulator and compared its performance to a conventional blockchain-based system. The simulation results show that Hedera achieved significantly lower latencies and better scalability than Ethereum, confirming its suitability for secure and efficient AI data verification in autonomous systems.
The rapid proliferation of Internet of Things (IoT) devices has introduced significant challenges in maintaining data security, integrity, and privacy, particularly in dynamic and multi-user environments where sensitive information is frequently exchanged across distributed networks. Conventional security mechanisms often fall short in addressing the unique demands of IoT systems due to their limited scalability, vulnerability to advanced cyber threats, and the need for real-time data processing. To address these concerns, this paper presents a novel and secure data-sharing architecture that seamlessly integrates Homomorphic Proxy Re-Encryption (HPRE) with blockchain technology. The proposed HPRE mechanism empowers a trusted proxy to re-encrypt ciphertext for designated recipients without accessing the original plaintext, thereby enabling flexible, privacy-preserving information exchange among authorized users. This allows IoT data owners to maintain control over their data while securely delegating access rights in a controlled and verifiable manner. To enhance accountability and eliminate single points of failure, a blockchain-based ledger is employed to immutably record all encryption operations, key-sharing events, and data access transactions. This integration guarantees transparency, traceability, and tamper-resistance, fostering a trustless environment suitable for highly sensitive IoT applications such as healthcare monitoring, industrial automation, and smart city infrastructures. Additionally, by incorporating lattice-based cryptography within the HPRE framework, the system achieves strong post-quantum security guarantees, rendering it resilient against both classical and emerging quantum computing threats. The proposed architecture is designed to be scalable, lightweight, and adaptable to heterogeneous IoT ecosystems, making it a reliable and future-proof solution for secure data exchange in privacy-critical applications. Experimental results and security analyses demonstrate the effectiveness and efficiency of the system in real-world IoT scenarios, confirming its potential for widespread adoption in next-generation secure IoT networks.
Caixiang Fan, Amirhossein Sohrabbeig, Petr MusıÌlek
Blockchain-based peer-to-peer energy trading enables individuals to directly share renewable energy using Internet of Things technologies. However, it faces significant challenges related to privacy, scalability, and the integration of advanced artificial intelligence. To address these issues, this article proposes zkPET, a secure and intelligent peer-to-peer energy trading framework. zkPET integrates machine learning and blockchain with advanced cryptographic techniques of zero-knowledge machine learning to protect user data while enabling intelligent decision making. In the zkPET framework, the computationally intensive operations of various machine learning models are executed off-chain, and only succinct cryptographic proofs of these computations are uploaded to the blockchain for verification and recording. In addition, a time-series clustering approach is incorporated into federated learning to enhance both inference accuracy and the efficiency of proof generation. Experimental validation using the zero-knowledge proof tool EZKL and a real-world electricity dataset demonstrates the feasibility and effectiveness of zkPET. The results underscore its potential to significantly improve privacy, scalability, and computational efficiency in decentralized energy trading, contributing to the advancement of secure and intelligent energy markets.
Diabetic Retinopathy (DR) detection in distributed telemedicine environments requires secure, scalable, and privacy-preserving solutions. Traditional federated learning (FL) relies on a central server, raising concerns about data privacy and system trust. We propose a novel serverless framework, FL-BC-SMPC-SMOTE, that integrates deep learning, FL, secure multi-party computation (SMPC), the Synthetic Minority Over-sampling Technique (SMOTE), Blockchain (Hyperledger Fabric), and the InterPlanetary File System (IPFS) to address these challenges. Using the APTOS 2019 dataset, we trained CNN-based models (e.g., EfficientNet-B0, ResNet-18) across 2â10 clients, achieving approximately 90% accuracy without raw data sharing. SMPC eliminates the need for a central aggregator by distributing encrypted model updates among clients, enabling privacy-preserving learning. Blockchain ensures auditable and tamper-resistant aggregation, while IPFS significantly reduces communication overheadâfrom 64 GB to 100 KB per round. Local SMOTE enhances recall for minority classes by 10â15%, promoting equity in DR severity classification. Compared to differentially private baselines (52.18% accuracy), our framework delivers a robust balance of performance, privacy, and fairness. This GDPR/HIPAA-compliant solution offers a practical and trustworthy approach to decentralized DR detection in real-world telemedicine settings.
Federated Learning (FL) has emerged as a critical paradigm for enabling privacy-preserving machine learning, particularly in regulated sectors such as finance and healthcare. However, standard FL strategies often encounter significant operational challenges related to fault tolerance, system resilience against concurrent client and server failures, and the provision of robust, verifiable privacy guarantees essential for handling sensitive data. These deficiencies can lead to training disruptions, data loss, compromised model integrity, and non-compliance with data protection regulations (e.g., GDPR, CCPA). This paper introduces Differentially Private Resilient Temporal Federated Learning (DP-RTFL), an advanced FL framework designed to ensure training continuity, precise state recovery, and strong data privacy. DP-RTFL integrates local Differential Privacy (LDP) at the client level with resilient temporal state management and integrity verification mechanisms, such as hash-based commitments (referred to as Zero-Knowledge Integrity Proofs or ZKIPs in this context). The framework is particularly suited for critical applications like credit risk assessment using sensitive financial data, aiming to be operationally robust, auditable, and scalable for enterprise AI deployments. The implementation of the DP-RTFL framework is available as open-source.