Sunil Vadera is the Dean of the School of
Computing, Science and Engineering at the University
of Salford in Greater Manchester, UK. Sunil was
Chair of the UK BCS Knowledge Discovery and Data
Mining Symposium held in Salford in 2009, A
Programme Chair of the IFIP conference on
Intelligent Information Processing in 2010,
2012,2016. His research has been published in some
of the leading outlets, including the Computer
Journal, ACM Transactions on Knowledge Discovery
from Data, ACM Computing Surveys, Expert Systems
Journal, Foundations of Science, and IEEE
Transactions of Power Systems. Sunil was Chair of
the British Computer Society Academic Accreditations
Committee, that has responsibility for professional
accreditation of programmes in the UK, from
2007-2009. He holds a PhD in Computer Science from
the University of Manchester, is a Fellow of the BCS
and was awarded the BDO best British Indian
Scientist and Engineer in 2014 and the Amity Award
for Research in 2018 in recognition of his
contributions to the field.
Sunil Vadera has led a number of projects in applying data mining and machine learning for problems in Energy, Finance, and Policy over the last decade, including:
• Developing new models for real time sensor validation of gas turbines
• Data mining of near miss data for the health and safety executive
• Analysis of SMART meters data for British Gas
• A major FP7 funded project on Self-Learning Energy Efficient Buildings and Open Spaces
• Analysing factors affecting children in need and troubled families
• Sub-prime lending aimed at improving financial inclusion
• Data mining for predicting client churn for a major Software House.
Abstract: Advances in deep learning have started a new AI revolution that is transforming our world. Applications in speech recognition, self-driving cars and publicity around systems such Alpha Go defeating human Go champions has ignited interest from the public, academia and industry. Convolutional neural networks, that can take images as input, learn to identify key features and perform classification are the heart of many of the proposed applications in medical diagnosis such as detecting breast cancer, predicting Alzheimer's disease and grading brain tumours. These neural networks can, however, be very large, taking up memory and requiring significant computational resources. For example, one of the most highly cited and early deep networks, AlexNet, has over 62 million parameters that need to be learned. This seminar explores methods for reducing the size of such networks without compromising performance. The talk will present a framework for categorising methods for pruning neural networks, covering magnitude-based pruning, sensitivity analysis methods, and use of clustering methods to identify similar weights. The talk will then present our recent work on a new framework based on the use of multi-armed bandits such as Thompson Sampling and Upper Confidence Bounds. It will conclude with the results of an empirical evaluation of the new methods over several benchmark data sets such as ImageNet, MNIST, CIFAR and Street View House Numbers.
Marat Akhmet is a professor of mathematics at Middle East Technical University (Ankara, Turkey) known for his research on the chaos and bifurcation theory in differential equations and hybrid systems with applications in physics, neural networks, biology, medicine and economics . Born in Kazakhstan, he studied at Aktobe State University. He received his doctorate in 1984 at Kiev University . He has been awarded a Science Prize of TUBITAK (Turkey, 2015), for best achievments in scientific research. He is an author of four books: "Principles of Discontinuous Dinamical Systems", Springer, 2010, "Nonlinear Hybrid Continuous Discrete-Time Models", Atlantis Press (Springer), 2011, "Neural networks with Discontinuous Impact Activations," Springer, 2014, and "Replication of Chaos in Neural Networks, Economics and Physics", Springer&HEP, 2015. His has solved the Second Peskin conjecture for Integrate-and-fire biological oscillators, has introduced and developed theory of differential equations with piecewise constant argument of generalized type, many aspects of discontinous dynamical systems. The last decade his main subject of research is input-output analysis of chaos and irregular behavior of hybrid neural networks.
Abstract: We aim to give arguments that chaos as a dynamical phenomenon and fractals as domains for chaos are generic objects in universe: natural as well as artificial. Specifically, genericity of chaos in neural networks dynamics and brain activity will be discussed. Moreover, we provide information on special tools, which extend chaos and fractals presence theoretically and in applications. For this reason, the abstractions are introduced and developed such as abstract similarity, abstract similarity map, abstract fractals and domain structured chaos.
Branislav Vuksanovic was born in Osijek, Croatia in
1962. He graduated from the University of Belgrade,
Serbia with degree in Electrical and Power
Engineering. He holds MSc degree in Measurement and
Instrumentation from South Bank University, London
and a PhD in Active Noise Control from the
University of Huddersfield, UK.
Previously, he worked as a Project Engineer for Croatian Electricity Board in Osijek, Croatia. During his academic career he worked as a Research Fellow at Sheffield and Birmingham Universities on Optical Brain Imaging and Medical Video Compression projects. He also worked as a Lecturer at the University of Derby where he was a member of Sensors and Controls Research Group. Currently he works as a Senior Lecturer at the University of Portsmouth, School of Engineering. He has published papers in the field of active noise control, biomedical signal processing and pattern recognition for intrusion detection and knowledge based authentication. He also published one book in Digital Electronics and Microcontrollers field.
Dr Branislav Vuksanovic is a member of IET, ILT and IACSIT. His current research interests are in the application of pattern recognition techniques for power systems and analysis of ground penetrating radar and ECG data.
Abstract: Facial expressions are the facial changes in response to a person’s internal emotional states, intentions, or social communications. Facial expression analysis has been an active research topic for behavioural scientists since the work of Darwin in 1872. Early attempts to automatically analyse facial expressions by tracking the motion of identified spots on an image sequence have started in the late 70s while the significant progress to build computer systems and techniques to help us understand and use this natural form of human communication has been made in more recent decades. In this talk, the problem of facial expression analysis will be considered in the context of computer systems and techniques designed to automatically analyse and recognize facial motions and facial feature changes from presented visual information. The first part of the presentation will discuss general structure of Automatic Facial Expression Analysis (AFEA) systems. The second part will describe the problem space for facial expression analysis including level of description, individual differences in subjects, transitions among expressions, intensity of facial expression, deliberate versus spontaneous expression, head orientation and scene complexity etc. Finally, some time will be devoted to a description of some specific approaches and techniques used in more recent works. They include the techniques for face acquisition, facial data extraction and representation, and facial expression recognition. Current status, future possibilities, and open questions about automatic facial expression analysis will be discussed at the end of the talk.