Federated learning (FL) offers a promising approach to training machine learning models on decentralized data sources without directly exchanging data. However, traditional differential privacy (DP) techniques, designed for centralized learning, often lead to substantial accuracy degradation in FL settings due to the inherent sensitivity of aggregated updates. This paper proposes a novel adaptive differential privacy (ADP) framework for FL that leverages quantization noise to mitigate this accuracy loss. Our approach dynamically adjusts the noise level based on the sensitivity of the local model updates, effectively reducing the overall noise while maintaining strong privacy guarantees. We demonstrate through theoretical analysis and simulations that our ADP framework achieves a superior trade-off between privacy and accuracy compared to standard DP approaches in FL. This work represents a significant advancement in the field by combining quantization and adaptive mechanisms for enhanced accuracy in federated learning.
Jincheng Zhang· Zenodo (CERN European Organi...· 0 citations
Federated learning (FL) offers a promising approach to train machine learning models on decentralized data sources without directly exchanging the data itself. However, the inherent privacy risks associated with FL remain a significant concern. Traditional differential privacy (DP) techniques often struggle to effectively protect privacy in FL settings, particularly when dealing with heterogeneous data distributions across participating devices. This paper proposes a novel framework for distributed differential privacy in federated learning that incorporates dynamic grouping of devices based on their data similarity and sensitivity. Our approach continuously analyzes data distributions across participating devices and adjusts grouping assignments to mitigate information leakage. We aim to achieve stronger differential privacy guarantees while maintaining reasonable data utility. The core of our method lies in a dynamic grouping algorithm which optimizes for both privacy and utility. We formally define the privacy loss and utility metrics, and provide a theoretical analysis of our framework. The key innovation is the adaptive grouping strategy, leading to more effective privacy protection.
Jincheng Zhang· Zenodo (CERN European Organi...· 0 citations
Federated learning (FL) offers a promising approach to training machine learning models on decentralized data sources without directly exchanging data. However, a key challenge in the widespread adoption of FL is the incentive problem: motivating participants to contribute their valuable data and computational resources. This paper proposes a novel decentralized federated learning system incorporating a blockchain-based incentive mechanism. The system leverages smart contracts on a blockchain to reward participants based on their contributions, enhancing the scalability, robustness, and trust of the FL process. The core claim is that incentivizing participation is a significant hurdle in FL, and this work presents a mechanism to address this. We detail the design of the system, focusing on the blockchain architecture, the smart contract implementation for reward distribution, and the overall protocol for federated learning. The system aims to create a self-regulating and trustworthy FL environment where participants are actively encouraged to contribute, ultimately leading to more robust and efficient model training.
Jincheng Zhang· Zenodo (CERN European Organi...· 0 citations
This paper proposes a novel decentralized federated learning (DFL) system designed to address the critical challenges of privacy and accountability inherent in collaborative learning environments. The system leverages the strengths of three key technologies: decentralized learning, differential privacy, and blockchain auditing. Traditional federated learning, while improving privacy by training models locally, still relies on a central server for aggregation, creating a single point of failure and potential vulnerability. This proposed architecture eliminates the central server, distributing the learning process across multiple participants. Differential privacy mechanisms are incorporated to further protect individual data contributions, adding noise to the model updates to obscure sensitive information. Finally, a blockchain-based auditing system provides a transparent and immutable record of all model updates, ensuring accountability and allowing for verification of the learning process. This combination offers a robust, secure, and auditable solution for distributed machine learning. The system's core claim is that ensuring privacy and accountability in federated learning is a major challenge, and this architecture directly addresses this need.
Jincheng Zhang· Zenodo (CERN European Organi...· 0 citations
Federated learning (FL) offers a promising approach to training machine learning models on decentralized data sources without directly sharing the data itself. However, existing FL systems face significant vulnerabilities, including malicious participant attacks and data breaches. This paper proposes a novel decentralized federated learning framework that integrates Byzantine fault tolerance (BFT) protocols, inspired by blockchain technology, with differential privacy (DP) mechanisms. The core claim is that this synergistic combination provides a substantially more robust and secure FL paradigm. Specifically, the system utilizes BFT to detect and mitigate malicious behavior from participants, ensuring data integrity and model robustness, while DP safeguards user privacy by adding controlled noise to the model updates. This framework allows the system to tolerate a certain number of Byzantine failures while maintaining accurate models. The system architecture is decentralized, eliminating a single point of failure and enhancing overall resilience. We demonstrate the potential of this approach through a theoretical analysis and outline a conceptual implementation strategy.
Jincheng Zhang· Zenodo (CERN European Organi...· 0 citations
Federated Learning (FL) offers a promising approach to training machine learning models on decentralized data sources without direct data sharing. However, the inherent lack of transparency in FL raises concerns about trust, fairness, and model reliability. This paper explores the critical need for explainable AI (XAI) techniques within the FL paradigm, specifically focusing on attributing contributions from individual participants to the global model. We propose a novel framework for tracing the influence of local model updates, quantifying the impact of each participant's contribution on the global model's parameters. Our methodology leverages gradient analysis and differential privacy considerations to provide a granular understanding of the learning process. The core claim is that understanding these contributions is essential for fostering trust and ensuring fairness in FL deployments. The developed attribution methods provide insights into the learning dynamics, allowing for the identification of potential biases or anomalous behavior. This work aims to establish a foundation for more robust and accountable FL systems.
Jincheng Zhang· Zenodo (CERN European Organi...· 0 citations
Federated learning (FL) offers a promising approach to training machine learning models on decentralized data sources, such as Internet of Things (IoT) devices, without directly exchanging raw data. However, the practical deployment of FL in IoT environments is hindered by several key challenges, including significant data heterogeneity across devices and the limited computational and storage resources inherent in these devices. This paper proposes a novel distributed learning framework that addresses these limitations by incorporating personalized experience replay (PER). PER allows each IoT device to store a representative subset of its training data and periodically replay past experiences to refine its local model. This mechanism not only mitigates the effects of data heterogeneity but also improves model convergence and accuracy, particularly in scenarios where individual devices have sparse or biased data. The proposed approach maintains user privacy by operating locally on device data, and the distributed nature of the framework reduces the communication overhead typically associated with FL. We demonstrate the effectiveness of our approach through a theoretical analysis and outline the key components and design considerations for a practical implementation. The core claim is that federated learning in IoT devices faces challenges with data heterogeneity and limited computational resources. The core mechanism is to implement a distributed learning framework incorporating personalized experience replay, where each IoT device stores a subset of its training data and periodically replays experiences to improve model performance, while preserving user privacy. This approach represents a significant advancement in FL for IoT, enabling more robust and accurate models in resource-constrained environments.
Jincheng Zhang· Zenodo (CERN European Organi...· 0 citations
Federated learning (FL) offers a promising approach to training machine learning models on decentralized data sources without directly exchanging data. However, traditional differential privacy (DP) techniques, designed for centralized learning, often lead to substantial accuracy degradation in FL settings due to the inherent sensitivity of aggregated updates. This paper proposes a novel adaptive differential privacy (ADP) framework for FL that leverages quantization noise to mitigate this accuracy loss. Our approach dynamically adjusts the noise level based on the sensitivity of the local model updates, effectively reducing the overall noise while maintaining strong privacy guarantees. We demonstrate through theoretical analysis and simulations that our ADP framework achieves a superior trade-off between privacy and accuracy compared to standard DP approaches in FL. This work represents a significant advancement in the field by combining quantization and adaptive mechanisms for enhanced accuracy in federated learning.
Jincheng Zhang· Zenodo (CERN European Organi...· 0 citations
Federated learning (FL) presents a promising paradigm for training machine learning models on decentralized data sources without directly exchanging the data itself. However, achieving differential privacy (DP) – a rigorous privacy guarantee – within the FL setting remains a significant challenge due to the computational overhead associated with traditional DP mechanisms. This work introduces a novel approach that combines differential privacy with secure multi-party computation (SMPC) to address this limitation. Our method employs a layered SMPC protocol to enable clients to perform gradient updates locally while maintaining privacy. The protocol minimizes data exposure by breaking down the computation into smaller, secure steps. The core claim is that existing DP mechanisms in FL are often computationally expensive. The proposed mechanism combines differential privacy with secure multi-party computation (SMPC) to perform gradient updates locally without revealing individual client data, utilizing a layered SMPC protocol for enhanced efficiency. This offers a practical and efficient solution for deploying DP in FL. The theoretical analysis demonstrates that our approach can achieve a desired privacy budget while significantly reducing the computational burden compared to standard DP techniques. This work contributes a new method to improve the performance of DP in FL.
Jincheng Zhang· Zenodo (CERN European Organi...· 0 citations
Federated learning (FL) offers a promising approach to training machine learning models on decentralized data without directly exchanging the data itself. However, traditional differential privacy (DP) methods often introduce substantial noise into the model updates, significantly degrading model accuracy. This paper proposes a novel adaptive noise injection scheme for FL, designed to mitigate this issue. The core idea is to dynamically adjust the noise level based on the sensitivity of the local data, thereby optimizing the privacy-utility trade-off. Our approach leverages the inherent variance in local updates to reduce the overall noise required to achieve a desired privacy guarantee. We demonstrate through theoretical analysis and intuitive simulations that our adaptive scheme leads to improved model accuracy compared to standard DP methods while maintaining the same level of privacy. The proposed method addresses a critical bottleneck in FL—the inherent conflict between privacy preservation and model performance—and provides a practical pathway for deploying robust and accurate FL systems.
Jincheng Zhang· Zenodo (CERN European Organi...· 0 citations
Federated learning (FL) offers a promising approach to train machine learning models on decentralized data sources without directly exchanging the data itself. However, the inherent privacy risks associated with FL remain a significant concern. Traditional differential privacy (DP) techniques often struggle to effectively protect privacy in FL settings, particularly when dealing with heterogeneous data distributions across participating devices. This paper proposes a novel framework for distributed differential privacy in federated learning that incorporates dynamic grouping of devices based on their data similarity and sensitivity. Our approach continuously analyzes data distributions across participating devices and adjusts grouping assignments to mitigate information leakage. We aim to achieve stronger differential privacy guarantees while maintaining reasonable data utility. The core of our method lies in a dynamic grouping algorithm which optimizes for both privacy and utility. We formally define the privacy loss and utility metrics, and provide a theoretical analysis of our framework. The key innovation is the adaptive grouping strategy, leading to more effective privacy protection.
Jincheng Zhang· Zenodo (CERN European Organi...· 0 citations
Federated learning (FL) offers a promising approach to training machine learning models on decentralized data without directly exchanging the data itself. However, traditional differential privacy (DP) methods often introduce substantial noise into the model updates, significantly degrading model accuracy. This paper proposes a novel adaptive noise injection scheme for FL, designed to mitigate this issue. The core idea is to dynamically adjust the noise level based on the sensitivity of the local data, thereby optimizing the privacy-utility trade-off. Our approach leverages the inherent variance in local updates to reduce the overall noise required to achieve a desired privacy guarantee. We demonstrate through theoretical analysis and intuitive simulations that our adaptive scheme leads to improved model accuracy compared to standard DP methods while maintaining the same level of privacy. The proposed method addresses a critical bottleneck in FL—the inherent conflict between privacy preservation and model performance—and provides a practical pathway for deploying robust and accurate FL systems.
Jincheng Zhang· Zenodo (CERN European Organi...· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
