2026· IEEE Transactions on Information Forensics and Security· Vol 21, pp. 7604-7618· 0 citations· 45 references
Abstract
Decision trees are widely used in various domains, such as user behavior analysis and financial risk assessment. Recently, increasing concerns about data privacy have driven the development of secure decision tree training frameworks. We propose ESecDT, a novel framework that enables collaborative decision tree training while preserving participants’ training data. ESecDT integrates the computational advantages of Function Secret Sharing (FSS) and Replicated Secret Sharing (RSS) through a co-design approach. We first introduce new protocols based on this co-design for tree training building blocks, including GroupSum, GroupPrefixSum, and VecMMax. Subsequently, we design a bit-width-aware training framework that manages data with different bit-widths and supports FSS key reuse. These designs enable ESecDT to ensure strong privacy guarantees, preserving the training data and all intermediate variables throughout the training process, while maintaining practical efficiency. Extensive experiments on nine real-world and synthetic datasets demonstrate that ESecDT reduces online communication overhead by <inline-formula> <tex-math notation="LaTeX">$4.25\times $ </tex-math></inline-formula>–<inline-formula> <tex-math notation="LaTeX">$5.56\times $ </tex-math></inline-formula> versus state-of-the-art frameworks. In the WAN setting, ESecDT achieves <inline-formula> <tex-math notation="LaTeX">$1.78\times $ </tex-math></inline-formula>–<inline-formula> <tex-math notation="LaTeX">$4.48\times $ </tex-math></inline-formula> speedup in training runtime and completes a 30,000-sample training task in less than half an hour, demonstrating strong potential for practical deployment.
As a classical type of machine learning algorithms, tree models have been widely employed in various fields, such as financial analysis and health diagnostics, offering high-accuracy and low-latency prediction services to users. However, tree evaluation also raises significant privacy concerns, particularly with respect to the tree model and the query sample, while the existing private decision tree evaluation schemes are unable to reach a good trade-off between privacy and efficiency in practice. Therefore, in this paper, we propose an efficient and privacy-preserving tree evaluation scheme based on additive homomorphic encryption, namely PACT. Specifically, PACT introduces an innovative algorithm by leveraging the overflow characteristic of two’s complement to support AHE-based parallel comparison, and it utilizes the lightweight homomorphic addition to select tree paths non-interactively. Meanwhile, we carefully design perturbation and shuffle methods to enhance model and sample privacy. The security of PACT is verified based on the ideal-real paradigm. Experimental results on real-world and synthetic datasets demonstrate the lossless accuracy and superior running efficiency of PACT.
Jiaqi Zhao, Hui Zhu, Junpeng Zhang et al.· IEEE Transactions on Informa...· 0 citations
It was concluded that strategies such as the careful selection of differential privacy parameters and training settings, along with the use of larger datasets, can improve the efficiency of FL and demonstrate that privacy-preserving and high-performance artificial intelligence systems can be securely applied in sensitive domains such as healthcare and finance.
Cagdas Karatas, Hibanur Karadogan, A. Ertug et al.· 0 citations
FHEON is presented, an open-source configurable framework for developing privacy-preserving neural network models for inference using the CKKS scheme of HE, and outperform all state-of-the-art HE inference works in both latency and memory utilization.
Nges Brian Njungle, Eric Jahns, Michel A. Kinsy· Proceedings on Privacy Enhan...· 1 citation
Federated learning (FL) enables collaborative model training across multiple clients in a privacy-preserving manner. However, the employment of homomorphic encryption algorithms might lead to high computational cost while the application of differential privacy (DP) methods would sacrifice model performance. To establish efficient and secure FL system as well as maintaining competitive performance, we introduce a DP-enabled cascaded filter with novel model-aggregation mechanism. Taking the model parameters of one client for example, dimensions with large absolute values and significant variations are selected by the cascaded filter and regarded as important dimensions. After that, random noise are added to these important dimensions for data security. Finally, the model parameters will be sent to the central server for aggregation. Unlike traditional DP-based approaches, our method considers each dimension’s informational importance, i.e., both absolute value and its variation. We theoretically prove the convergence of our method and verify the effectiveness on different scenarios including four datasets. The experimental results suggest that our method outperforms the other ones from literature under both IID and non-IID conditions while achieving high level of efficiency and privacy protection performance.
Zhiqiang Chen, Yuchen Jiang, Ray Y. Zhong et al.· IEEE Transactions on Informa...· 0 citations
Decision tree classification serves as a fundamental component in many machine learning applications. As inference services are increasingly outsourced to cloud platforms, designing privacy-preserving mechanisms has become important. To secure outsourced decision tree inference, Chen et al. proposed SecDT, an efficient secret-sharing-based framework. While SecDT offers notable efficiency and preliminary security guarantees, we identify a critical vulnerability: attribute leakage. Specifically, the attributes associated with decision tree nodes are exposed to non-owner parties, which risks revealing sensitive model information. In this paper, we first present two secure enhancements, SecDT+v1 and SecDT+v2. These variants use the transformation matrix technique to obfuscate node attributes within the secret-shared domain. To prevent more advanced leakage through attribute access patterns, we design two advanced variants, SecDT+vH and SecDT+vDP, which incorporate dot-product operations to achieve the attribute-hiding property. Extensive evaluations on real-world datasets demonstrate that our proposed schemes provide robust security guarantees while outperforming state-of-the-art solutions in both classification latency and bandwidth efficiency.
Chang-Ching Hung, Tsung-Che Hsieh, Yu-Chi Chen et al.· IEEE Transactions on Informa...· 0 citations
This paper comprehensively reviews the application of Federated Learning in fingerprint recognition, a distributed machine learning paradigm that enables collaborative modeling without sharing raw data, thus offering a promising privacy-preserving solution.