Compact and Efficient Neural Networks for Image Recognition Based on Learned 2D Separable Transform

TL;DR

This paper introduces a learned 2D separable transform (LST) layer that reduces model complexity and parameters in neural networks for image recognition, achieving high accuracy with efficient implementation.

Contribution

The paper proposes a novel LST layer that shares weights across rows and columns, significantly decreasing parameters while maintaining high recognition accuracy.

Findings

Achieved 98.02% accuracy on MNIST with only 9.5k parameters.

Demonstrated FPGA implementation for efficient, compact neural network deployment.

LST-based models outperform traditional stacked FC layers in parameter efficiency.

Abstract

The paper presents a learned two-dimensional separable transform (LST) that can be considered as a new type of computational layer for constructing neural network (NN) architecture for image recognition tasks. The LST based on the idea of sharing the weights of one fullyconnected (FC) layer to process all rows of an image. After that, a second shared FC layer is used to process all columns of image representation obtained from the first layer. The use of LST layers in a NN architecture significantly reduces the number of model parameters compared to models that use stacked FC layers. We show that a NN-classifier based on a single LST layer followed by an FC layer achieves 98.02\% accuracy on the MNIST dataset, while having only 9.5k parameters. We also implemented a LST-based classifier for handwritten digit recognition on the FPGA platform to demonstrate the efficiency of the suggested approach for designing a compact and high-performance implementation of NN models. Git repository with supplementary materials: https://github.com/Mak-Sim/LST-2d