CoShNet: A Hybrid Complex Valued Neural Network using Shearlets

TL;DR

CoShNet introduces a hybrid neural network using shearlets for fixed transforms, achieving high accuracy with significantly fewer parameters and FLOPs compared to traditional CNNs, and requiring less training time without hyperparameter tuning.

Contribution

The paper presents a novel hybrid neural network using shearlets as fixed transforms, improving feature representation and efficiency over wavelet-based methods.

Findings

Achieves 92.2% accuracy on Fashion-MNIST, outperforming ResNet-50 and ResNet-18.

Uses 52 times fewer FLOPs and significantly fewer parameters than ResNet-18.

Requires only 20 epochs for training without hyperparameter tuning.

Abstract

In a hybrid neural network, the expensive convolutional layers are replaced by a non-trainable fixed transform with a great reduction in parameters. In previous works, good results were obtained by replacing the convolutions with wavelets. However, wavelet based hybrid network inherited wavelet's lack of vanishing moments along curves and its axis-bias. We propose to use Shearlets with its robust support for important image features like edges, ridges and blobs. The resulting network is called Complex Shearlets Network (CoShNet). It was tested on Fashion-MNIST against ResNet-50 and Resnet-18, obtaining 92.2% versus 90.7% and 91.8% respectively. The proposed network has 49.9k parameters versus ResNet-18 with 11.18m and use 52 times fewer FLOPs. Finally, we trained in under 20 epochs versus 200 epochs required by ResNet and do not need any hyperparameter tuning nor regularization. Code: https://github.com/Ujjawal-K-Panchal/coshnet