notesum.ai

Published at December 10

Post-Training Non-Uniform Quantization for Convolutional Neural Networks

cs.CV

cs.LG

Released Date: December 10, 2024

Authors: Ahmed Luqman¹, Khuzemah Qazi¹, Imdadullah Khan¹

Aff.: ¹Lahore University of Management Sciences

Arxiv: http://arxiv.org/pdf/2412.07391v1

Refer to caption

DNN Model	FP32 Top-1 Acc	Precision	Top 1 Accuracy on ImageNet
			Ours	Uniform	APoT
ResNet18	69.97	4 bits	46.16	44.18	47.82
		5 bits	55.88	52.03	52.13
		6 bits	59.14	52.49	51.98
		7 bits	61.41	52.64	52.66
		8 bits	62.53	52.41	52.83