CipherFormer: Efficient Transformer Private Inference with Low Round Complexity

被引:0
|
作者
Wang, Weize [1 ]
Kuang, Yi [1 ]
机构
[1] Shanghai Jiao Tong Univ, Sch Cyber Sci & Engn, Shanghai, Peoples R China
来源
PROCEEDINGS OF THE 2024 27 TH INTERNATIONAL CONFERENCE ON COMPUTER SUPPORTED COOPERATIVE WORK IN DESIGN, CSCWD 2024 | 2024年
关键词
Private Inference; Transformer; Homomorphic Encryption; Garbled Circuit;
D O I
10.1109/CSCWD61410.2024.10580655
中图分类号
TP39 [计算机的应用];
学科分类号
081203 ; 0835 ;
摘要
There is a growing trend to outsource the inference task of large transformer models to cloud servers. However, this poses a severe threat to users' private data as they are exposed to cloud servers after uploading. Although several works attempted to provide private inference for transformer models, their hundreds of communication rounds limit the application scenarios. Motivated by the desire to minimize round complexity, we propose CipherFormer, a novel transformer private inference scheme using homomorphic encryption and garbled circuits. We present a protocol for quickly computing homomorphic matrix multiplications. We then modify the attention mechanism and design the corresponding garbled circuits. Furthermore, we show how to use a lightweight attention mechanism and mixed-bitwidth to reduce the inference latency while maintaining accuracy. In comparison with an advanced homomorphic encryption scheme on text classification tasks, our model improves accuracy by 3% to 11% while performing private inference with a 7.7x-11.9x speedup.
引用
收藏
页码:3054 / 3059
页数:6
相关论文
共 50 条
  • [31] Towards Efficient Vision Transformer Inference: A First Study of Transformers on Mobile Devices
    Wang, Xudong
    Zhang, Li Lyna
    Wang, Yang
    Yang, Mao
    PROCEEDINGS OF THE 2022 THE 23RD ANNUAL INTERNATIONAL WORKSHOP ON MOBILE COMPUTING SYSTEMS AND APPLICATIONS (HOTMOBILE '22), 2022, : 1 - 7
  • [32] Communication-Efficient Model Parallelism for Distributed In-Situ Transformer Inference
    Wei, Yuanxin
    Ye, Shengyuan
    Jiang, Jiazhi
    Chen, Xu
    Huang, Dan
    Du, Jiangsu
    Lu, Yutong
    2024 DESIGN, AUTOMATION & TEST IN EUROPE CONFERENCE & EXHIBITION, DATE, 2024,
  • [33] Efficient Transformer Inference for Extremely Weak Edge Devices using Masked Autoencoders
    Liu, Tao
    Li, Peng
    Gu, Yu
    Liu, Peng
    ICC 2023-IEEE INTERNATIONAL CONFERENCE ON COMMUNICATIONS, 2023, : 1718 - 1723
  • [34] Embedding-Free Transformer with Inference Spatial Reduction for Efficient Semantic Segmentation
    Yu, Hyunwoo
    Cho, Yubin
    Kang, Beoungwoo
    Moon, Seunghun
    Kong, Kyeongbo
    Kane, Suk-Ju
    COMPUTER VISION - ECCV 2024, PT XLII, 2025, 15100 : 92 - 110
  • [35] On the Complexity of Differentially Private Data Release Efficient Algorithms and Hardness Results
    Dwork, Cynthia
    Naor, Moni
    Reingold, Omer
    Rothblum, Guy N.
    Vadhan, Salil
    STOC'09: PROCEEDINGS OF THE 2009 ACM SYMPOSIUM ON THEORY OF COMPUTING, 2009, : 381 - 390
  • [36] Efficient block entropy coding with low complexity
    Said, A
    Pearlman, WA
    Islam, A
    1998 IEEE INTERNATIONAL SYMPOSIUM ON INFORMATION THEORY - PROCEEDINGS, 1998, : 139 - 139
  • [37] Falcon: Accelerating Homomorphically Encrypted Convolutions for Efficient Private Mobile Network Inference
    Xu, Tianshi
    Li, Meng
    Wang, Runsheng
    Huang, Ru
    2023 IEEE/ACM INTERNATIONAL CONFERENCE ON COMPUTER AIDED DESIGN, ICCAD, 2023,
  • [38] An Efficient and Low-Complexity Transformer-Based Deep Learning Framework for High-Dynamic-Range Image Reconstruction
    Lopez-Cabrejos, Josue
    Paixao, Thuanne
    Alvarez, Ana Beatriz
    Luque, Diodomiro Baldomero
    SENSORS, 2025, 25 (05)
  • [39] PRIVCORE: Multiplication-activation co-reduction for efficient private inference
    Pang, Zhi
    Wang, Lina
    Yu, Fangchao
    Zhao, Kai
    Zeng, Bo
    Xu, Shuwang
    NEURAL NETWORKS, 2025, 187
  • [40] Low Complexity In-Loop Filter for VVC Based on Convolution and Transformer
    Feng, Zhen
    Jung, Cheolkon
    Zhang, Hao
    Liu, Yang
    Li, Ming
    IEEE ACCESS, 2024, 12 : 120316 - 120325