Keynote Lectures

The Electroencephalogram (EEG) Analysis and Classification for Diagnosis and Prognosis of Brain Disorders
António Dourado, University of Coimbra, Portugal, Portugal

Towards Lifelong Learning in Optimisation Algorithms
Emma Hart, Edinburgh Napier University, United Kingdom, United Kingdom

Ambient Intelligent Systems: The Role of Non-Intrusive and Sensitive Approaches
Paulo Novais, Universidade do Minho, Portugal, Portugal

Type-2 Fuzzy Systems for Human Decision Making
Jonathan Garibaldi, University of Nottingham, United Kingdom, United Kingdom

Abstract
The Electroencephalogram (EEG) is the most used multidimensional biosignal  in the study of the brain condition. Brain disorders such as depression, epilepsy, Alzheimer, Parkinson, dementia, etc., have influence in the EEG signals, allowing to process them as biomarkers of these diseases, for diagnosis and prognosis, if the proper features are found and extracted from the EEG. A brief review of the state of the art concerning the EEG acquisition and processing will be discussed. Generally the biomarking is made as a classification problem following the usual stages : pre-processing, features selection and extraction, classification, post-processing. The main difficulties in this activity concerning EEG and other biosignals is the nonlinear, nonstationary nature of the signals, and the high variability of them from patient to patient, and even for the same patient. This reduces the generalization capabilities of the classifiers, which in most of the cases use computational intelligence tools, such as artificial neural networks and support vector machines. An overview of the research going on concerning these problems will be presented and discussed, with particular emphasis in the prediction and detection of epileptic seizures.

Abstract
Optimisation is an important activity for many businesses, providing better, faster, cheaper solutions to problems in areas including scheduling of people and processes, routing of vehicles and  packing of containers. Metaheuristic algorithms provide a pragmatic way to tackle optimisation, providing high-quality solutions in reasonable time. Unfortunately, selection and tuning of an appropriate algorithm can difficult, often requiring an expert to design the algorithm, a software engineer to implement it, and finally application of automated tuning processes to refine the chosen algorithm.  This is not only costly, requiring significant human-effort, but also results in software which can quickly become obsolete when it no longer matches the goals of a company or if the characteristics of the optimisation problems being solved changed substantially. Unlike human-beings, optimisation software is currently unable to adapt to changing scenarios or autonomously improve its behaviour over time as it learns from experience.
To counter this, I will propose the life-long learning optimisation system (L2O) which when faced with a continual stream of problems to optimise,  refines an existing set of algorithms so that they improve over time as they are exposed to more examples,  and automatically generates new algorithms when faced with problem instances that are completely different from those seen before. The approach is inspired by ideas from the operation of the natural immune system, which exhibits many properties of a life-long learning system that can be exploited computationally, and uses genetic programming to automatically generate new algorithms. I will give a brief overview of the immune system, focusing on highlighting its relevant computational properties and then show how it can be used to construct a lifelong learning optimisation system. The system is shown to adapt to new problems, exhibit memory, and produce efficient and effective solutions when tested in both the bin-packing and scheduling domains, representing a paradigm shift in the way we think about optimisation.

Abstract
There is currently a significant interest in consumer electronics in applications and devices that monitor and improve the user’s well-being. This is one of the key aspects in the development of ambient intelligence systems. Nonetheless, existing approaches are generally based on physiological sensors, which are intrusive and cannot be realistically used, especially in ambient intelligence in which the transparency, pervasiveness and sensitivity are paramount. We put forward a new approach to the problem in which user behavioral cues are used as an input to assess inner state. This innovative approach has been validated by research in the last years and has characteristics that may enable the development of true unobtrusive, pervasive and sensitive ambient intelligent systems.

Abstract
Type-2 fuzzy sets and systems, including both interval and general type-2 sets, are now firmly established as tools for the fuzzy researcher that may be deployed on a wide range of applications and in a wide set of contexts. However, in many situations the output of type-2 systems are type-reduced and then defuzzified to an interval centroid, which are then often even simply averaged to obtain a single crisp output. Many successful applications of type-2 have been in control contexts, often focussing on reducing the RMSE. This is not taking full advantage of the extra modelling capabilities inherent in type-2 fuzzy sets. In this talk, I will present some of the current research being carried out within the LUCID group at Nottingham, and wider, into type-2 for modelling human reasoning. I will cover approaches and methodologies which make more use of type-2 capabilities, illustrating these with reference to practical applications such as classification of breast cancer tumours, modelling expert variability in cyber-security contexts, and other decision support problems.