AE-Qdrop: Towards Accurate and Efficient Low-Bit Post-Training Quantization for A Convolutional Neural Network

AE-Qdrop: Towards Accurate and Efficient Low-Bit Post-Training Quantization for A Convolutional Neural Network

Blockwise reconstruction with adaptive rounding helps achieve acceptable 4-bit post-training quantization accuracy. However, adaptive rounding is time intensive, and the optimization space of weight elements is constrained to a binary set, thus limiting the performance of quantized models. The optim...

Full description

Bibliographic Details
Main Authors: Jixing Li, Gang Chen, Min Jin, Wenyu Mao, Huaxiang Lu
Format: Article
Language:English
Published: MDPI AG 2024-02-01
Series:Electronics
Subjects:
post-training quantization
adaptive rounding
block-wise reconstruction
progressive optimization strategy
randomly weighted quantized activation
global fine-tuning
Online Access:https://www.mdpi.com/2079-9292/13/3/644

Internet

https://www.mdpi.com/2079-9292/13/3/644

Similar Items