Investigating Efficient Compression Models to Accelerate Deep Neural Networks

In recent years, owing to the remarkable progress in technology and computer power, artificial intelligence models have gained a widespread attention. In fact, these artificial models are being produced and used to solve a large number of current and future business and technical problems.Nowadays, Deep Neural Networks are becoming more and more involved in different fields. Implementing such deep networks on embedded devices is remaining an important challenging task considering the massive requirement of computation and storage. Thus, Deep Learning researchers have come up with a large number of powerful optimization techniques to accelerate deep models.One of these techniques is model compression, which involves reducing neural network’s size while preserving its accuracy. This decrease in size holds significance as larger networks pose challenges when it comes to deployment on devices having restricted capabilities.In this work, we will explore the strengths and weaknesses of some well-known compression methods and present our new approach to accelerate deep networks using a compression model based on the Wavelet Beta without compromising accuracy.In fact, when compared to widely recognized compression models, our new approach enhances the efficient outcomes and performances demonstrated in this paper within the realm of image compression.