Numeric image classification with TensorFlow

Recognition of alphanumeric data using a machine learning algorithm is a problem with practical applications in license plate, traffic sign, and street number recognition. TensorFlow, an open source software library originally developed by the Google Brain Team, offers a flexible architecture and an easy to learn interface that allows for rapid implementation of and evaluation of different machine learning algorithms and data structures. This paper covers predictive analysis of the SVHN, or Street View House Numbers, dataset using a Convolution Neural Network model developed on the TensorFlow platform. The goal of the paper is to improve the training speed and validation accuracy of an existing CIFAR-10 neural network model implemented in TensorFlow by changing its activation functions, regularization measures, number of convolution layers, loss optimizers, and architectural organization.