Evolutionary fine tuning of quantized convolution-based deep learning models

TL;DR

This paper introduces an evolutionary optimization method to enhance the accuracy of quantized deep learning models, addressing the limitations of nearest neighbor quantization in pretrained networks.

Contribution

It proposes using evolution strategies to improve quantization accuracy by selectively adjusting weights, outperforming traditional nearest neighbor methods.

Findings

Evolution strategies can significantly improve quantized model accuracy.

The approach is effective on architectures like VGG, ResNet, and autoencoders.

Results show faster accuracy improvements compared to standard quantization.

Abstract

Deep learning models are the most efficient models in many machine learning tasks. The main disadvantage when using them in IoT, mobile devices, independent autonomous or real-time systems is their complexity and memory size. Therefore, much research has concentrated on compression techniques of deep learning architectures. One of the most popular technique is quantization. In most of the works, the quantization is done based on the nearest neighbour quantization technique. This work focuses on improving the quantization efficiency in pretrained and quantized models. This approach has the potential to improve the final accuracy of quantized models. The main postulate of the work is that final quantization states of the network based on nearest neighbour rounding does not guarantee optimal accuracy. In the presented work, the evolution strategy is used as an optimization approach. The evolution in each iteration changes the values of the small percentage of weights. It shifts theirs values to different quantization states. The work shows that proposed evolution with an appropriate set of operators and parameters can fast improve the accuracy of the quantized models. The results are presented for popular architectures such as VGG and Resnet for image classification and detection. Additionally, simulations were carried out for the autoencoder architecture.