Improving the Efficiency of Transformers for Resource-Constrained Devices

被引:12
|
作者
Tabani, Hamid [1 ]
Balasubramaniam, Ajay [1 ]
Marzban, Shabbir [1 ]
Arani, Elahe [1 ]
Zonooz, Bahram [1 ]
机构
[1] NavInfo Europe, Adv Res Lab, Eindhoven, Netherlands
关键词
Deep Learning; Transformers; Clustering; Resource-Constrained Devices;
D O I
10.1109/DSD53832.2021.00074
中图分类号
TP3 [计算技术、计算机技术];
学科分类号
0812 ;
摘要
Transformers provide promising accuracy and have become popular and used in various domains such as natural language processing and computer vision. However, due to their massive number of model parameters, memory and computation requirements, they are not suitable for resource-constrained low-power devices. Even with high-performance and specialized devices, the memory bandwidth can become a performance-limiting bottleneck. In this paper, we present a performance analysis of state-of-the-art vision transformers on several devices. We propose to reduce the overall memory footprint and memory transfers by clustering the model parameters. We show that by using only 64 clusters to represent model parameters, it is possible to reduce the data transfer from the main memory by more than 4x, achieve up to 22% speedup and 39% energy savings on mobile devices with less than 0.1% accuracy loss.
引用
收藏
页码:449 / 456
页数:8
相关论文
共 50 条
  • [41] Distributed Symbolic Network Quality Assessment for Resource-constrained Devices
    Augello, Andrea
    Gaglio, Salvatore
    Lo Re, Giuseppe
    Peri, Daniele
    2021 26TH IEEE INTERNATIONAL CONFERENCE ON EMERGING TECHNOLOGIES AND FACTORY AUTOMATION (ETFA), 2021,
  • [42] Secure and Low-Power Authentication for Resource-Constrained Devices
    Sethi, Mohit
    Kortoci, Pranvera
    Di Francesco, Mario
    Aura, Tuomas
    PROCEEDINGS 2015 5TH INTERNATIONAL CONFERENCE ON THE INTERNET OF THINGS (IOT), 2015, : 30 - 36
  • [43] Low Latency Implementations of CNN for Resource-Constrained IoT Devices
    Mujtaba, Ahmed
    Lee, Wai-Kong
    Hwang, Seong Oun
    IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS II-EXPRESS BRIEFS, 2022, 69 (12) : 5124 - 5128
  • [44] Iterative neural networks for adaptive inference on resource-constrained devices
    Sam Leroux
    Tim Verbelen
    Pieter Simoens
    Bart Dhoedt
    Neural Computing and Applications, 2022, 34 : 10321 - 10336
  • [45] Iterative neural networks for adaptive inference on resource-constrained devices
    Leroux, Sam
    Verbelen, Tim
    Simoens, Pieter
    Dhoedt, Bart
    NEURAL COMPUTING & APPLICATIONS, 2022, 34 (13): : 10321 - 10336
  • [46] Unobtrusive Occupancy Detection with FastGRNN on Resource-Constrained BLE Devices
    Billah, Md Fazlay Rabbi Masum
    Campbell, Bradford
    PROCEEDINGS OF THE 1ST ACMWORKSHOP ON DEVICE-FREE HUMAN SENSING (DFHS 19), 2019, : 1 - 5
  • [47] A Review of Lightweight Security and Privacy for Resource-Constrained IoT Devices
    Kumar, Sunil
    Kumar, Dilip
    Dangi, Ramraj
    Choudhary, Gaurav
    Dragoni, Nicola
    You, Ilsun
    CMC-COMPUTERS MATERIALS & CONTINUA, 2024, 78 (01): : 31 - 63
  • [48] Extending ZigBee Tree Routing Protocol for Resource-Constrained Devices
    Roy, Uttam Kumar
    2014 IEEE ASIA PACIFIC CONFERENCE ON WIRELESS AND MOBILE, 2014, : 48 - 53
  • [49] FuzzyKey: Comparing Fuzzy Cryptographic Primitives on Resource-Constrained Devices
    Zhang, Mo
    Marin, Eduard
    Oswald, David
    Singelee, Dave
    SMART CARD RESEARCH AND ADVANCED APPLICATIONS (CARDIS 2021), 2022, 13173 : 289 - 309
  • [50] Lightweight Stream Cipher Scheme for Resource-Constrained IoT Devices
    Noura, Hassan
    Couturier, Raphael
    Pham, Congduc
    Chehab, Ali
    2019 INTERNATIONAL CONFERENCE ON WIRELESS AND MOBILE COMPUTING, NETWORKING AND COMMUNICATIONS (WIMOB), 2019,