Private Decision Tree Evaluation with Malicious Security via Function Secret Sharing

Jiaxuan Fu, Ke Cheng, Yuheng Xia, Anxiao Song, Qianxing Li, Yulong Shen

Abstract: Private Decision Tree Evaluation (PDTE) allows the client to use a decision tree classification model on the server to classify their private data, without revealing the data or the classification results to the server. Recent advancements in PDTE have greatly enhanced its effectiveness in scenarios involving semi-honest security, offering a viable and secure alternative to traditional, less secure approaches for decision tree evaluation. However, this model of semi-honest security may not always align well with real-world problems. In this work, we present FSSTree, a malicious-secure three-party PDTE protocol using function secret sharing (FSS). FSSTree achieves its high performance against malicious adversaries via several innovative cryptographic designs. Especially, 1) we transform a comparison operation into a prefix parity query problem, allowing us to implement malicious-secure comparisons rapidly using lightweight and verifiable FSS. 2) Building upon this, we further propose a constant-round protocol for securely evaluating Conditional Oblivious Selection (COS). 3) We utilize these optimized protocols to enhance the PDTE processes, achieving a considerable decrease in both communication costs and the number of rounds. The experimental results show that FSSTree reduces the runtime in a WAN environment by up to \(22.3\times \) times and saves up to \(4.1 \times \) the communication cost compared to the state-of-the-art work.

External IDs:dblp:conf/esorics/FuCXSLS24