Challenges and Solutions in Developing Convolutional Neural Networks and Long Short-Term Memory Networks for Industry Problems

Abstract

Developing a deep learning application requires unique approaches and methods. These are discussed in two industrial applications. The architectures used are two classic ones: convolutional neural network (CNN) and long short-term memory (LSTM). The CNN application to identify hand-written choice options from a scanned image, faced issues of “pose” (oriental transformation) problem and “region segmentation.” The LSTM application to predict the equated monthly instalment by all customers had to overcome challenges of “multi-variate” and “multiple time series.” The evolution of a satisfactory solution for the customer in both these cases is described in this chapter. The importance of data analysis, data pre-processing and proper choice of hyper-parameters is seen through these experiments.