The ISC International Journal of Information Security

Abstract Federated learning enables collaborative model training without centralized data collection, but existing frameworks rely on a central server, introducing risks of single points of failure, adversarial manipulation, and privacy leakage. To address these challenges, we propose HashLearner, a secure decentralized learning framework that utilizes the HashGraph consensus protocol for model aggregation without trusted authorities. HashLearner introduces two key innovations: (i) a consensus-driven decentralized aggregation mechanism resilient to Byzantine adversaries, and (ii) a privacy-preserving shuffling strategy that mitigates gradient reconstruction and poisoning attacks. To handle heterogeneous data distributions, the framework further employs transfer learning–based personalization. The simulation results of HashLearner, tested on benchmark Kaggle datasets, demonstrate that the platform maintains high accuracy while significantly enhancing scalability, security, and privacy. These findings indicate that HashLearner provides a practical path toward scalable, privacy-preserving, and trustworthy decentralized federated learning.

Abstract Adaptive data analysis (ADA) involves a dynamic interaction between an analyst and a dataset owner, where the analyst submits queries sequentially, adapting them based on previous answers. This process can become adversarial, as the analyst may attempt to overfit by targeting non-generalizable patterns in the data. To counteract this, the dataset owner introduces randomization techniques, such as adding noise to the responses. This noise not only helps prevent overfitting, but also enhances data privacy. However, it must be carefully calibrated to ensure that the statistical reliability of the responses is not compromised. In this paper, we extend the ADA problem to the context of distributed datasets. Specifically, we consider a scenario where a potentially adversarial analyst interacts with multiple distributed responders through adaptive queries. We assume the responses are subject to noise, introduced by the channel connecting the responders and the analyst. We demonstrate how this noise can be opportunistically leveraged through a federated mechanism to enhance the generalizability of ADA, thereby increasing the number of query-response interactions between the analyst and the responders. We illustrate that the careful tuning of the transmission amplitude based on the theoretically achievable bounds can significantly impact the number of accurately answerable queries.

Abstract In a federated learning system, the objective is to train a global model over distributed datasets without centralizing all data on a single unit. This is accomplished by training a local model on the dataset of each data owner and then aggregating these local models to preserve the datasets’ privacy. To incentivize clients to actively engage in the learning process, fairness-aware federated learning techniques can be employed. One such approach involves quantifying the contribution of locally trained models in training the global model by Shapley value (SV) using an additional dataset and rewarding them according to their contributions. However, the calculation of the Shapley value presents a significant challenge due to its high computational complexity. To tackle this issue, our research presents a contribution-based federated learning method that efficiently computes the contribution of each locally trained model by distributing the additional dataset among processing nodes in a private manner and calculating the Shapley value over them.

Abstract Large-scale data collection is challenging in alternative centralized learning as privacy concerns or prohibitive policies may rise. As a solution, Federated Learning (FL) is proposed wherein data owners, called participants, can train a common model collaboratively while their privacy is preserved. However, recent attacks, namely Membership Inference Attacks (MIA) or Poisoning Attacks (PA), can threaten the privacy and performance in FL systems. This paper develops an innovative Adversarial-Resilient Privacy-preserving Scheme (ARPS) for FL to cope with preceding threats using differential privacy and
cryptography. Our experiments display that ARPS can establish a private model with high accuracy out‌performing state-of-the-art approaches. To the best of our knowledge, this work is the only scheme providing privacy protection beyond any output models in conjunction with Byzantine resiliency without sacrificing accuracy and efficiency.