EGL-B mathematics conference11 June 2019
EGL-B workshop summary
The EGL-B mathematics conference begins on 4th of June with an introduction from Professor Sabah Jassim of the University of Buckingham, welcoming the speakers and attendees from the universities of Essex, Greenwich, London and Buckingham. The topics discussed enabled individuals to gain an insight and the chance to present their research on Optimisation, Applied and Numerical Mathematics and Data Analysis. These sessions also incorporated the mathematics of Artificial Intelligence, Machine Learning in Big Data analysis and applications.
Following the introduction to the EGL-B mathematics workshop, day one began with the theme of “Applied Optimisation”, followed by Automating Computational Models and Artificial Intelligence and Machine Learning Strategies.
Students who presented a seminar during the workshop were able to gain constructive feedback to their presentations and research by other university representors. Three students, Mr Fakher Mohammed (University of Buckingham), Ms Gayatri Rode (University of Buckingham) and Mr Adam Woodhouse (University of Essex) were awarded certificates for a high quality presentation with elaborate research at the EGL-B workshop along with a gift card.
Professor Sabah Jassim presentation summary
Professor Sabah Jassim begins his talk by introducing Big Data and what it means in the emergence of social media, networking and how deep learning is attracting a high interest in computer applications. He explains the purpose of Machine Learning and the requirements to utilise it’s features once extracted. Leading on to describe the features used in data, how the texture and structural features in image domains are extensively extracted and fed into AI algorithms.
Sabah then defines what Machine Learning classifiers are and the objectives of Deep Learning by comparing the convolutional layers to one another. He shows how Convolutional Neural Network (CNN) has been described to use several convolutional layers to learn large numbers by discriminating features using random linear filters followed by a non linear option.
As the presentation continues, he explains the advantages of using Deep Learning by referring to its performance, reducing the need for advance searching with known discriminating power, adaptability to new related problems and varying architectures which are widely available. This is followed by the Deep Learning disadvantages which are related to a large requirement for training data, overfitting classifications, expensive to train and test samples and may misclassify testing samples dissimilar to the training samples.
Sabah refers to the notable attempts in overcoming Deep Learning disadvantages by using Tensorflow and transfer learning to extend the applicability to cases where samples are lacking. Topological data analysis is also used as a supplement with similar themes being available to Deep Learning. Explaining how it would be possible to use CS convolutional layers and by using CS based random projections, this may reduce the amount of redundancies generated and may help reduce overfitting.
It was an eventful two days for the EGL-B mathematics workshop and the School of Computing was delighted to be able to host this year’s event. A special thank you to all the staff and students who attend and presented on behalf of the universities of Buckingham, Essex, London and Greenwich, with attendees from Kent and Middlesex also taking part.