Current approaches to explainable AI (XAI) frequently fall short of providing genuinely understandable and trustworthy explanations. These methods often rely on post-hoc interpretations of black-box models, which can be misleading due to their failure to capture the underlying causal mechanisms driving predictions. This paper proposes a novel axiomatic foundation for XAI, arguing that true explanation necessitates a thorough understanding of causal relationships. We posit that causal inference represents a fundamental requirement for explainability, moving beyond superficial feature importance or local linear approximations. This framework introduces a rigorous approach to assessing explanations by evaluating their consistency with known causal structures, leading to more robust and reliable explanations. We define key concepts and provide a formal outline for evaluating explanation methods based on their ability to accurately represent and leverage causal knowledge. The core contribution is establishing a clear criterion – causal fidelity – for evaluating XAI methods, ensuring explanations reflect the true causal drivers of model behavior.
Jincheng Zhang· Zenodo (CERN European Organi...· 0 citations
Federated learning (FL) offers a promising approach to training machine learning models across decentralized devices while preserving data privacy. However, FL systems are vulnerable to Byzantine faults, where malicious participants introduce corrupted model updates, compromising the integrity of the global model. This paper proposes a novel framework for distributed federated learning with Byzantine fault tolerance, leveraging the power of homomorphic encryption. The system encrypts model updates using homomorphic encryption, allowing the central server to perform aggregation computations directly on the encrypted data. This approach eliminates the need for decryption, thereby safeguarding client data and providing robust protection against Byzantine attacks. The core claim of this work is that the combination of FL with homomorphic encryption provides a fundamentally secure and resilient solution for distributed model training. We detail the system architecture, the encryption and aggregation protocols, and analyze the security and performance implications. The proposed method significantly enhances the robustness of FL against malicious participants, offering a practical solution for privacy-sensitive applications.
Jincheng Zhang· Zenodo (CERN European Organi...· 0 citations
This paper presents a novel framework for distributed learning of Generative Adversarial Networks (GANs) through a federated learning approach. Traditional GAN training demands centralized computation and large-scale datasets, posing significant challenges for privacy-sensitive and resource-constrained environments. This research addresses these limitations by enabling collaborative GAN training across multiple decentralized devices, termed as Federated Generative Adversarial Networks (FedGANs). The core claim is that training GANs across distributed devices presents substantial challenges, and this paper proposes a framework to overcome these challenges. The framework leverages federated learning to allow devices to generate data collaboratively without sharing raw data, thereby preserving data privacy. The design focuses on maintaining the integrity and diversity of generated samples. This work introduces a new approach to federated learning specifically optimized for GANs, contributing to the advancement of privacy-preserving and decentralized generative modeling. The proposed system is evaluated conceptually, outlining the key components and potential improvements.
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, the inherent data heterogeneity and potential biases present in these datasets pose significant challenges to achieving algorithmic fairness. This paper investigates the application of randomized smoothing, a technique traditionally used in batch learning, to mitigate bias and improve fairness in federated learning settings. We propose a novel framework that adapts randomized smoothing to the asynchronous and decentralized nature of FL. The core idea is to introduce controlled noise during model aggregation, effectively averaging out biased updates from clients with differing data distributions. Our approach demonstrates improved fairness metrics compared to standard federated averaging, particularly in scenarios with significant data skew. We formally define the problem, outline the methodology, and discuss the theoretical implications of our findings. This work represents a new direction in fairness-aware federated learning and contributes to the development of more robust and equitable machine learning systems.
Jincheng Zhang· Zenodo (CERN European Organi...· 0 citations
Federated learning (FL) offers a promising paradigm for training machine learning models on decentralized data sources without directly exchanging the data itself. However, traditional FL approaches often suffer from significant communication overhead and potential privacy vulnerabilities. This research proposes a novel scalable federated learning system that addresses these limitations through the integration of graph-based communication topologies and differential privacy mechanisms. The core idea is to represent the relationships between participating clients as a graph, enabling selective communication and aggregation of model updates based on the graph structure, thereby reducing unnecessary communication and enhancing privacy. The system utilizes a graph convolutional network (GCN) to learn client embeddings reflecting their connectivity and data characteristics, which then guide the communication process. Differential privacy is incorporated by adding calibrated noise to the aggregated model updates, providing a provable privacy guarantee. Simulation results demonstrate that the proposed approach achieves significant improvements in communication efficiency and privacy compared to standard FL methods, particularly in scenarios with complex client relationships and stringent privacy requirements. This work contributes a practical and scalable solution for deploying federated learning in diverse applications where data privacy and communication efficiency are paramount.
Jincheng Zhang· Zenodo (CERN European Organi...· 0 citations
Federated learning (FL) offers a promising paradigm for training machine learning models on decentralized data sources without directly exchanging data. However, achieving strong differential privacy (DP) guarantees in FL is a significant challenge, particularly when combined with compression techniques commonly employed to reduce communication costs. Traditional approaches often rely on fixed compression ratios, potentially sacrificing accuracy for privacy or vice versa. This paper introduces an adaptive differential privacy framework for FL that dynamically adjusts the compression ratio based on the sensitivity of the data and the desired privacy level. We leverage randomized compression methods, specifically noise addition with learned parameters, to achieve this adaptation. Our approach demonstrates a novel balance between privacy and utility, offering a more effective solution compared to static compression strategies. The core claim of this work is that achieving strong differential privacy guarantees in federated learning necessitates careful management of compression ratios. The key mechanism involves introducing an adaptive compression scheme. This paper presents a formalization of this adaptive approach, outlining its mathematical foundations and providing a framework for its implementation.
Jincheng Zhang· Zenodo (CERN European Organi...· 0 citations
Federated learning (FL) offers a promising paradigm for training machine learning models on decentralized data sources without directly exchanging the data itself. However, traditional FL approaches often suffer from significant communication overhead and potential privacy vulnerabilities. This research proposes a novel scalable federated learning system that addresses these limitations through the integration of graph-based communication topologies and differential privacy mechanisms. The core idea is to represent the relationships between participating clients as a graph, enabling selective communication and aggregation of model updates based on the graph structure, thereby reducing unnecessary communication and enhancing privacy. The system utilizes a graph convolutional network (GCN) to learn client embeddings reflecting their connectivity and data characteristics, which then guide the communication process. Differential privacy is incorporated by adding calibrated noise to the aggregated model updates, providing a provable privacy guarantee. Simulation results demonstrate that the proposed approach achieves significant improvements in communication efficiency and privacy compared to standard FL methods, particularly in scenarios with complex client relationships and stringent privacy requirements. This work contributes a practical and scalable solution for deploying federated learning in diverse applications where data privacy and communication efficiency are paramount.
Jincheng Zhang· Zenodo (CERN European Organi...· 0 citations
Federated learning (FL) presents a promising approach to training machine learning models across decentralized devices while preserving data privacy. However, traditional FL methods often lack sufficient privacy guarantees, leaving users vulnerable to privacy breaches. This work proposes a novel framework for distributed federated learning incorporating differential privacy (DP). Our approach enhances the standard FL process by adding noise to the model updates, effectively masking individual user contributions. This guarantees a rigorous privacy budget, quantified by the ε and δ parameters of the DP mechanism. The resulting system enables personalized model training – adapting the global model to individual user data – while maintaining strong privacy protections. We outline the core components of the system, including client selection, local model training with DP noise injection, and aggregation of noisy updates. The theoretical analysis demonstrates the effectiveness of the proposed method in achieving strong privacy guarantees alongside reasonable model accuracy. The presented framework provides a viable path for building trust-worthy and privacy-preserving collaborative learning systems.
Jincheng Zhang· Zenodo (CERN European Organi...· 0 citations
Federated learning (FL) offers a promising paradigm for training machine learning models across decentralized devices while preserving data privacy. However, recent research has demonstrated that FL systems are susceptible to adversarial attacks, where maliciously crafted inputs can significantly degrade model performance. This paper addresses the critical issue of adversarial robustness in FL by proposing a novel approach combining randomized smoothing and formal verification. We leverage randomized smoothing to generate robust models and then utilize formal verification techniques to provide certified robustness guarantees against adversarial perturbations. Our method ensures that the trained FL model will maintain a certain level of accuracy even when subjected to adversarial attacks within a specified bounded range. The key contribution lies in the synergistic integration of these two approaches, offering a practical and theoretically sound solution for building secure and reliable FL systems. We demonstrate the effectiveness of our approach through a detailed analysis and theoretical framework, establishing a foundation for future research in this area.
Jincheng Zhang· Zenodo (CERN European Organi...· 0 citations
Predictive maintenance aims to anticipate equipment failures, minimizing downtime and maintenance costs. Traditional approaches often rely on centralized data collection, raising significant privacy concerns, particularly when dealing with sensitive sensor data from industrial equipment. This research proposes a novel framework that leverages the strengths of federated learning and edge-based anomaly detection to achieve robust and privacy-preserving predictive maintenance. The core idea is to train a predictive maintenance model collaboratively across multiple devices using federated learning, while simultaneously employing local anomaly detection models residing on edge devices to identify deviations from normal operation *before* data transmission. This approach reduces the amount of raw data sent to a central server, thereby mitigating privacy risks and potentially improving model accuracy through localized anomaly detection. The framework is designed to be scalable and adaptable to various industrial environments. We present a detailed discussion of the system architecture, the federated learning process, and the edge-based anomaly detection techniques. The proposed method offers a compelling solution for industries seeking to proactively manage equipment health while safeguarding sensitive data.
Jincheng Zhang· Zenodo (CERN European Organi...· 0 citations
Federated learning (FL) offers a promising approach to training machine learning models across decentralized devices while preserving data privacy. However, FL systems are vulnerable to Byzantine faults, where malicious participants introduce corrupted model updates, compromising the integrity of the global model. This paper proposes a novel framework for distributed federated learning with Byzantine fault tolerance, leveraging the power of homomorphic encryption. The system encrypts model updates using homomorphic encryption, allowing the central server to perform aggregation computations directly on the encrypted data. This approach eliminates the need for decryption, thereby safeguarding client data and providing robust protection against Byzantine attacks. The core claim of this work is that the combination of FL with homomorphic encryption provides a fundamentally secure and resilient solution for distributed model training. We detail the system architecture, the encryption and aggregation protocols, and analyze the security and performance implications. The proposed method significantly enhances the robustness of FL against malicious participants, offering a practical solution for privacy-sensitive applications.
Jincheng Zhang· Zenodo (CERN European Organi...· 0 citations
This paper proposes a novel distributed Bayesian optimization framework utilizing federated learning to optimize black-box functions across multiple clients while preserving data privacy. The core concept involves each client independently training a local Bayesian model based on its local dataset. Subsequently, a central server orchestrates the optimization process by aggregating these local models through a federated averaging algorithm, thereby updating a global Bayesian model. This approach eliminates the need for direct data sharing, a critical advantage in scenarios where data privacy is paramount. We demonstrate the efficacy of this framework, highlighting its potential for applications in areas such as hyperparameter tuning, robotics, and drug discovery, where data is often distributed and sensitive. The theoretical foundation rests on Bayesian optimization principles and the established methodologies of federated learning, creating a robust and adaptable solution. This work contributes to the growing field of privacy-preserving optimization.
Jincheng Zhang· Zenodo (CERN European Organi...· 0 citations