Tutorial IV

Abstract - The evolution of artificial neural networks as mathematical models emanating from findings from the field of neuroscience has been a key and significant thread in AI research. The application of deep learning has taken centre stage, and with advances in processor architecture and speed, memory capacity and computational algorithms, their ubiquitous presence in intelligent devices is expected to grow further. A crucial aspect of an artificial neural network remains the chosen architecture and with it, the associated parameters used to configure the individual neurons and the training process used to adjust the underlying parameters. This workshop aims to make a brief historical chronology of the evolution of artificial neural networks, using the Perceptron first to illustrate the simplest neural network, yet effective in solving certain classes of classification problems, and from there, the appeal and challenges on connecting them to form a MultiLayer Perceptron (MLP). The role of dimensionality in deep learning is illustrated using Principal Component Analysis as a factor reduction technique, while polynomial expansion can be used for the opposite. Used in conjunction, PCA and polynomial expansion is demonstrated as a powerful combination for enhancing discrimination between classes. Finally, the ability to preserve the rich neighbourhood information available in grid data as opposed to the single dimensionality of MLP datasets is demonstrated through application of Convolutional Neural Networks for processing images with case studies in the medical and surveillance fields.

Dr. Mahendra Gooroochurn works as Senior Lecturer in the Mechanical & Production Engineering Department of the University of Mauritius. He is a Chartered Engineer registered with the Engineering Council of UK, an accredited green building design and construction professional registered with the USGBC, an Edge Expert registered with the International Finance Corporation (IFC) and holds the following professional engineering memberships: MIET, MIEEE and MASHRAE. He has over three years industry experience as research manager and head of sustainability in building services consultancy and has research interests in sustainability in the built environment, circular design, circular economy and community engagement for climate action. He has special interest in applying AI in developing innovative solutions in his research and is a certified Huawei Instructor for AI from the ICT Academy at the University of Mauritius.