ArXiv TLDR

GPU Acceleration of Sparse Fully Homomorphic Encrypted DNNs

🐦 Tweet
2604.11659

Lara D'Agata, Carlos Agulló-Domingo, Óscar Vera-López, Kaustubh Shivdikar, Ardhi W. B. Yudha + 5 more

cs.CRcs.DCcs.DScs.LGcs.PF

TLDR

This paper accelerates sparse fully homomorphic encrypted deep neural networks on AMD GPUs, improving ciphertext matrix multiplication runtime and complexity.

Key contributions

  • Targets FHE matrix multiplication, a computationally intensive DNN operation.
  • Proposes an optimized method for ciphertext matmul using FIDESlib on AMD GPUs.
  • Exploits sparsity in operands to improve performance and reduce time complexity.
  • Achieves up to 3.0x speedup over CPU and reduces complexity from cubic to semi-linear.

Why it matters

Fully homomorphic encryption is vital for secure computation, particularly in machine learning. Accelerating FHE operations on GPUs significantly advances privacy-preserving AI, making secure DNNs more practical. This work demonstrates a path towards efficient, secure deep learning.

Original Abstract

Fully homomorphic encryption (FHE) has recently attracted significant attention as both a cryptographic primitive and a systems challenge. Given the latest advances in accelerated computing, FHE presents a promising opportunity for progress, with applications ranging from machine learning to information security. We target the most computationally intensive operation in deep neural networks from a hardware perspective, matrix multiplication (matmul), and adapt it for execution on AMD GPUs. We propose a new optimized method that improves the runtime and complexity of ciphertext matmul by using FIDESlib, a recent open-source FHE library designed specifically for GPUs. By exploiting sparsity in both operands, our sparse matmul implementation outperforms its CPU counterpart by up to $3.0\times$ and reduces the time complexity from cubic to semi-linear, demonstrating an improvement over existing FHE matmul implementations.

📬 Weekly AI Paper Digest

Get the top 10 AI/ML arXiv papers from the week — summarized, scored, and delivered to your inbox every Monday.