paperbot · PL 论文追踪

全部论文 待读 收藏

P2 GPU-based Private Information Retrieval for On-Device Machine Learning Inference

面向端侧机器学习推理的基于 GPU 的私有信息检索
ASPLOS Proceedings of the 29th ACM International Conference on Architectural Support for Programming Languages and Operating Systems, Volume 12024匹配度 1/10 · 该论文研究 GPU 加速的私有信息检索及与端侧 ML 推理的协同设计，属于隐私保护、系统与体系结构方向，与用户关注的静态分析、形式化验证、浮点/贝叶斯程序分析以及 LLM 与 PL 的交叉均无直接关系。引用 24 inbox
Maximilian Lam, Jeff Johnson, Wenjie Xiong, Kiwan Maeng, Udit Gupta, Yang Li, Liangzhen Lai, Ilias Leontiadis, Minsoo Rhu, Hsien-Hsin S. Lee, Vijay Janapa Reddi, Gu-Yeon Wei, David Brooks, Edward Suh
安全与隐私语言与体系结构
crossref: 2026-09-12dblp: 2026-09-12arxiv: 2026-09-12

AI 速览

一句话：提出 GPU 加速的 PIR 并与下游 ML 协同设计，将端侧推荐/语言模型推理的嵌入检索吞吐提升百倍且保持精度。

端侧机器学习推理可保护用户隐私，但推荐模型等依赖 1–10 GB 的嵌入表，难以全部存放在设备上。论文提出用私有信息检索（PIR）从服务器秘密获取嵌入，避免泄露查询信息；针对现成 PIR 计算开销过大，提出 GPU 加速的 PIR 实现，并与下游 ML 应用协同设计以进一步降低延迟。评估显示，相比优化的 CPU PIR，GPU 加速使吞吐提升超过 20×，PIR-ML 协同设计在固定模型质量下再提升超过 5×；单块 V100 可达每秒 10 万次查询，较 CPU 基线提升超过 100×，且保持模型精度。

问题
端侧私有 ML 推理需要访问过大而无法本地存储的嵌入表，直接使用 PIR 又因计算密集难以满足延迟敏感推理。
方法
论文提出 GPU 加速的 PIR 算法实现，并与下游 ML 应用进行协同设计，以减少嵌入检索与推理开销。
评估
在推荐与语言建模等端侧 ML 任务上，与优化 CPU PIR 相比吞吐提升超过 20×，协同设计再提升超过 5×；单块 V100 达 10 万 QPS，较 CPU 基线超过 100× 且保持精度。
原文摘要（Abstract）

On-device machine learning (ML) inference can enable the use of private user data on user devices without revealing them to remote servers. However, a pure on-device solution to private ML inference is impractical for many applications that rely on embedding tables that are too large to be stored on-device. In particular, recommendation models typically use multiple embedding tables each on the order of 1--10 GBs of data, making them impractical to store on-device. To overcome this barrier, we propose the use of private information retrieval (PIR) to efficiently and privately retrieve embeddings from servers without sharing any private information. As off-the-shelf PIR algorithms are usually too computationally intensive to directly use for latency-sensitive inference tasks, we 1) propose novel GPU-based acceleration of PIR, and 2) co-design PIR with the downstream ML application to obtain further speedup. Our GPU acceleration strategy improves system throughput by more than 20× over an optimized CPU PIR implementation, and our PIR-ML co-design provides an over 5× additional throughput improvement at fixed model quality. Together, for various on-device ML applications such as recommendation and language modeling, our system on a single V100 GPU can serve up to 100,000 queries per second---a > 100× throughput improvement over a CPU-based baseline---while maintaining model accuracy.

链接与引用

DOI 原文 · arXiv · PDF（开放获取） · DBLP

BibTeX 
@inproceedings{LamJ0MGLLLRLRW024,
  title = {GPU-based Private Information Retrieval for On-Device Machine Learning Inference},
  author = {Maximilian Lam and Jeff Johnson and Wenjie Xiong and Kiwan Maeng and Udit Gupta and Yang Li and Liangzhen Lai and Ilias Leontiadis and Minsoo Rhu and Hsien-Hsin S. Lee and Vijay Janapa Reddi and Gu-Yeon Wei and David Brooks and Edward Suh},
  booktitle = {ASPLOS},
  year = {2024},
  doi = {10.1145/3617232.3624855}
}