A Layer-Wise Theoretical Framework for Deep Learning of Convolutional Neural Networks

As research attention in deep learning has been focusing on pushing empirical results to a higher peak, remarkable progress made the performance race of machine applications past years. Yet based artificial neural networks still remains difficult understand as it is considered black-box approach. A lack understanding from theoretical perspective would not only hinder employment them where high-stakes decisions need be made, but also limit their future development intelligence expected robust, predictable and trustable. This paper aims provide methodology investigate train convolutional so ensure convergence. mathematical model matrix representations for first formulated an analytic layer-wise framework then proposed tested several common benchmarking image datasets. The case studies show reasonable trade-off between accuracy learning, highlight potential employing method finding appropriate number layers actual implementations.