notesum.ai

Published at December 5

Quantized and Interpretable Learning Scheme for Deep Neural Networks in Classification Task

cs.LG

cs.CV

Released Date: December 5, 2024

Authors: Alireza Maleki¹, Mahsa Lavaei², Mohsen Bagheritabar³, Salar Beigzad⁴, Zahra Abadi⁵

Aff.: ¹Department of Technology Management University of Tehran; ²Electrical and Computer Engineering University of Tehran; ³Electrical Engineering Department, University of Cincinnati; ⁴Software and Engineering University of St. Thomas, Minnesota; ⁵Electrical Engineering Department Tehran University

Arxiv: http://arxiv.org/pdf/2412.03915v1

Hyperparameter	MNIST	CIFAR-10
Learning Rate	0.1	0.01
Epochs	50	50
Batch Size	128	64
Quantization Bits ( $k$ )	8	4
Clipping Parameter ( $\alpha$ )	Learnable	Learnable
Saliency Masking Ratio	50%	50%
Regularization Term ( $\lambda$ )	0.1	0.05