EP-HDC: Hyperdimensional Computing with Encrypted Parameters for High-Throughput Privacy-Preserving Inference

TL;DR

This paper addresses privacy-preserving inference for high-throughput applications by combining hyperdimensional computing with encrypted parameters (EP-HDC) and implementing client-side HE. By encrypting class hypervectors and performing similarity search on the client, EP-HDC removes the need to encrypt query hypervectors, dramatically reducing encryption and communication overhead while maintaining accuracy. Extensive design-space exploration (quantization, architecture, HE parameters) and BFV/BGV-aware strategies yield substantial throughput and latency improvements—up to hundreds of times faster than prior HE-HDC and DNN-based PPML methods with less than 1% accuracy loss. The results indicate EP-HDC as a practical, scalable PPML approach for batch inference in privacy-sensitive settings.

Abstract

While homomorphic encryption (HE) provides strong privacy protection, its high computational cost has restricted its application to simple tasks. Recently, hyperdimensional computing (HDC) applied to HE has shown promising performance for privacy-preserving machine learning (PPML). However, when applied to more realistic scenarios such as batch inference, the HDC-based HE has still very high compute time as well as high encryption and data transmission overheads. To address this problem, we propose HDC with encrypted parameters (EP-HDC), which is a novel PPML approach featuring client-side HE, i.e., inference is performed on a client using a homomorphically encrypted model. Our EP-HDC can effectively mitigate the encryption and data transmission overhead, as well as providing high scalability with many clients while providing strong protection for user data and model parameters. In addition to application examples for our client-side PPML, we also present design space exploration involving quantization, architecture, and HE-related parameters. Our experimental results using the BFV scheme and the Face/Emotion datasets demonstrate that our method can improve throughput and latency of batch inference by orders of magnitude over previous PPML methods (36.52~1068x and 6.45~733x, respectively) with less than 1% accuracy degradation.

Figures (4)

• Figure 1: HE-HDC vs. the proposed EP-HDC for batch HDC inference. HV stands for hypervector.
• Figure 2: Encryption method comparison between BFV, BGV, and CKKS.
• Figure 3: Relationship among EP-HDC parameters.
• Figure 4: $D_{\max}$ for various $N$ and $\log_2 t$. The width of shaded region represents standard deviation.