Eureka! Institutional Repository

This thesis aims at presenting platforms, tools and techniques for the training of distributed Machine Learning models through Google's Tensorflow and comparing it with the competitors of this platform and the resources they provide to facilitate datascience technology engineers and programmers. In addition, the terms Machine Learning and Distributed Training are analyzed for their role in today's industries. We also explore the ways and techniques of exporting a trained model of learning mechanics to use and implement it on Mobile Platforms on Android and iOS. Then we showcase the training of a distributed model in Tensorflow, through Python and its libraries. This model is trained on the MNIST dataset to create a system that can recognize numeric digits from 0 to 9, by taking it as an input of a greyscale 28x28 image. We then process this model and convert it into the form it needs to be in order to be utilised, from the corresponding frameworks of Android and iOS. Finally, we create the applications that allow a user to draw a numeric digit from 0 to 9 and by pressing a button the screen displays the predicted outcome of the model we previously imported. The thesis was conducted by Moumoulides Alexandros and Tzoumas Christos, students of Alexander Technological Institute of Thessaloniki and was supervised by professor Mr. Diamantaras Konstantinos