ProgressiveNN: Achieving Computational Scalability with Dynamic Bit-Precision Adjustment by MSB-first Accumulative Computation
-
- Suzuki Junnosuke
- Tokyo Institute of Technology
-
- Kaneko Tomohiro
- Tokyo Institute of Technology
-
- Ando Kota
- Tokyo Institute of Technology
-
- Hirose Kazutoshi
- Tokyo Institute of Technology
-
- Kawamura Kazushi
- Tokyo Institute of Technology
-
- Chu Thiem Van
- Tokyo Institute of Technology
-
- Motomura Masato
- Tokyo Institute of Technology
-
- Yu Jaehoon
- Tokyo Institute of Technology
Abstract
Computational scalability allows neural networks on embedded systems to provide desirable inference performance while satisfying severe power consumption and computational resource constraints. This paper presents a simple yet scalable inference method called ProgressiveNN, consisting of bitwise binary (BWB) quantization, accumulative bit-serial (ABS) inference, and batch normalization (BN) retraining. ProgressiveNN does not require any network structure modification and obtains the network parameters from a single training. BWB quantization decomposes and transforms each parameter into a bitwise format for ABS inference, which then utilizes the parameters in the most-significant-bit-first order, enabling progressive inference. The evaluation result shows that the proposed method provides computational scalability from 12.5% to 100% for ResNet18 on CIFAR-10/100 with a single set of network parameters. It also shows that BN retraining suppresses accuracy degradation of training performed with low computational cost and restores inference accuracy to 65% at 1-bit width inference. This paper also presents a method to dynamically adjust the bit-precision of the ProgressiveNN to achieve a better trade-off between computational resource use and accuracy for practical applications using sequential data with proximity resemblance. The evaluation result indicates that the accuracy increases by 1.3% with an average bit-length of 2 compared with only the 2-bit BWB network.
Journal
-
- International Journal of Networking and Computing
-
International Journal of Networking and Computing 11 (2), 338-353, 2021
IJNC Editorial Committee
- Tweet
Keywords
Details 詳細情報について
-
- CRID
- 1390288624491299456
-
- NII Article ID
- 130008063326
-
- ISSN
- 21852847
- 21852839
-
- Text Lang
- en
-
- Data Source
-
- JaLC
- Crossref
- CiNii Articles
- KAKEN
-
- Abstract License Flag
- Allowed