Main supervisor: Professor Amir Hussain

Malignant melanoma (i.e. skin cancer) is the third most frequent type of skin cancer and one of the most malignant tumors, accounting for 79% of skin cancer deaths. Despite its aggressive infiltration of other body parts, it is highly curable if diagnosed early and treated properly.

The standard approach in automatic dermoscopic image analysis has usually three stages: i) image segmentation; ii) feature extraction and feature selection; and iii) lesion classification. The segmentation stage is considered the most important stage since it affects the accuracy of the subsequent steps. However, segmentation is difficult because of the great variety of lesion shapes, size, and colors along with different skin types and textures.

Deep Learning, currently a hot research topic, is considered the leading machine learning tool in the imaging and computer vision domains. It can be seen as an improvement to the artificial neural networks as it consists of more layers. In other words, Deep Learning constructs many layers of abstraction that help map inputs to higher level representation.

The ultimate goal of applying machine learning to medical images is to recognize patterns in a better and quicker way than humans can, and thus increasing the productivity of doctors and the patient healthcare outcomes. Deep Learning (a form of machine learning) is able to take that further, especially its ability to provide improved predictions from the large amount of data it is trained on, due to the higher levels of abstractions it provides. We believe that Deep Learning could effectively contribute to the early detection of melanoma, and facilitate in distinguishing between benign and malignant moles.

Project Website: http://www.deepderma.io

Automatic recognition of human emotions has been a hot topic of research over the last decade, with a vast number of published papers exploiting several modalities, such as face, physiological signals including electrocardiogram (ECG) and Electroencephalography (EEG), gesture, postures, pupillary dilation and speech. The last mentioned modality can be considered as one of the most targeted in terms of human emotion recognition. The interest of engineers in developing systems capable of recognizing human’s emotions from speech can be explained by the unlimited applications of such systems in many fields, such as robotics, telecommunications services, e-learning among many others. The aim of the doctoral thesis is to propose a system capable of recognizing man emotional state with a high reliability, in real conditions and in real time. The proposed and realized system should be robust to environmental noise (this includes the acoustic noise and rapid body movements). The research will be also devoted to some very important aspects, including the discussion of differences between optimal extracted features from each modality depending on the age, gender and cultural background of subjects.

The aim of this project is developing computational models for real world applications such as audio-visual integration in hearing-aids, sentic computing, autonomous vehicles, etc by emulating the functionality of the corresponding region in the brain. Starting with the Hippocampus in the brain which is responsible to spatial navigation and episodic memory, currently, the project is focussed on three areas (1) trying to generalize the recurrent collaterals model to other regions of the brain which have recurrent collaterals; (2) working on audio-visual integration emulation from the working mechanisms of polysensory neurons in the superior colliculus. (3) working on emotion diagnosis in sentiment analysis by studying the region in the brain responsible for identifying emotions which is amygdala.

Papers from this work, so far:
Shiva. A, Hussain. A, "Continuous Time Recurrent Neural Network Model of Recurrent Collaterals in the Hippocampus CA3 Region", BICS 2016 conference, Beijing.

Cloud elasticity enables the dynamic re-adjustment of the underlying computational resources thus facilitating clients to meet their performance demands and paying only for the resources that are used. However, determining the right amount of resources that maintains the system performance at a desired level whilst minimizing the running cost is a challenging task. This becomes more difficult, when the systems deployed over cloud are subject to operate in time-varying and unpredictable operating conditions. This project aims to explore bio-inspired methods to develop a new intelligent cloud elasticity framework for systems that operate in time varying and unpredictable operating conditions. The project focus on multiple techniques including multi-model/multiple feedback control architecture, fuzzy logic, bio-inspired/brain-inspired computational models and multi-objective optimization for the development of framework.   

Researcher: Muhammad Saad Razzaq

Clustering is one of the most challenging optimization problems. Many Swarm Intelligence techniques including Ant Colony optimization (ACO), Particle Swarm Optimization (PSO), and Honey Bee Optimization (HBO) have been used to solve clustering. Cat Swarm Optimization (CSO) is one of the newly proposed heuristics in swarm intelligence, which is generated by observing the behavior of cats and has been used for clustering and numerical function optimization. CSO based clustering is dependent on a pre-specified value of K i.e. Number of Clusters. In this paper, we have proposed a “Modified Cat Swam Optimization (MCSO)” heuristic to discover clusters based on the nature of data rather than user specified K. We have compared the results of MCSO with CSO to demonstrate the enhanced efficiency and accuracy of our proposed technique

Researcher: Fahad Maqbool

The Graph Coloring Problem (GCP) is about assigning colors (labels) to the vertices of a graph by using minimum number of colors, such that no two vertices joined by an edge have the same color. In this paper we have proposed a technique BNMR-COL for GCP based on Blind Naked Mole-Rats (BNMR) Meta heuristic. A new local move has been devised to generate neighboring solutions for GCP, which is further used in modeling the proposed BNMR-COL. Experiments were carried out using standard benchmark graph instances. Comparative results of proposed algorithm with results of previously proposed algorithms were reported.

Customer churn is a critical and challenging problem affecting business and industry, in particular, the rapidly growing, highly competitive telecommunication sector. It is of substantial interest to both academic researchers and industrial practitioners, interested in forecasting the behavior of customers in order to differentiate the churn from non-churn customers. The primary motivation is the dire need of businesses to retain existing customers, coupled with the high cost associated with acquiring new ones. A review of the field has revealed a lack of efficient, rule-based Customer Churn Prediction (CCP) approaches in the telecommunication sector. This study proposes an intelligent rule-based decision-making technique, based on rough set theory (RST), to extract important decision rules related to customer churn and non-churn. The proposed approach effectively performs classification of churn from non-churn customers, along with prediction of those customers who will churn or may possibly churn in the near future. Extensive simulation experiments are carried out to evaluate the performance of our proposed RST based CPP approach using four rule-generation mechanisms, namely, the Exhaustive Algorithm (EA), Genetic Algorithm (GA), Covering Algorithm (CA) and the LEM2 algorithm. Empirical results show that RST-GA is the most efficient technique for extracting implicit knowledge in the form of decision rules from the publicly available, benchmark telecom dataset. Further, comparative results demonstrate that our proposed approach offers a globally optimal solution for CCP in the telecom sector, when benchmarked against several state-of-the-art methods. Finally, we show how attribute-level analysis can pave the way for developing a successful customer retention policy, that could form an indispensable part of strategic decision making and planning process in the telecom sector.

Researcher: Dr. Tahseen Jilani

Data Clustering is a process of organization of data into groups such that objects in a group are similar to each other and dissimilar with objects in other groups. Data clustering algorithms have been used for a broad range of problems, including historic data analysis, image segmentation, and financial markets analysis for portfolio management. In the recent past, there is also increase in the use of nature inspired algorithms for the use of data clustering to solve many real-world optimization problems. Granular Computing (GC) is also a rising model of information processing that deals with handling pieces of information, called information granules. In this paper, a fuzzy clustering method is used with Particle Swarm Optimization (PSO) using Granule Computing to develop and design a method, for stock market portfolio management. For experiments, we have used financial data of companies listed on the Hong Kong Stock Exchange. The result shows that Fuzzy Particle Swarm Optimization (FPSO) based clustering with Granule Computing builds efficient clusters for portfolio management. The returns of portfolios designed with this method are compared with the portfolio returns with the Hong Kong Stock Exchange benchmark index, i.e. Hang Sang Composite Index (HSCI).  The results of portfolios designed using our method are better than the benchmark index HSCI.

In normal cell condition, most of the genes are not expressed. These genes could be altered by certain biological stage or cell development while others are septic to tissue types or environmental conditions. Elucidating the hidden patterns in gene expression profiles offers a tremendous opportunity for an enhanced and in-depth understanding of functional genomics. Monitoring gene expression levels in different developmental stages, tissue types, clinical conditions, and different organisms can help understand gene functions and gene networks, assist in the diagnosis of disease conditions and reveal the effects of medical treatments. The prospection of these hidden patterns in gene expression data for the prognosis and diagnosis of disease especially cancer is of a great interest and promising results. This research aims at combining the flow of data, not only from RNA (and protein synthesis) but ensemble the data from epigenetics into an intelligent decision support system.

Researcher: Kia Dashtipour

The overall aim is to research, design and develop a novel customer-centric, multi-modal emotion-sensitive human-computer interface, and a range of intelligent features that can add value and significantly enhance the functionality of an existing Web/Mobile App based wedding services management system.

Web 2.0 plays a major in our daily lives. We express our emotions and we state our opinion toward a specific topic or a certain product. These sentiments and emotions could enhance us in making a decision or give the opportunity to compare between different products or different companies. We not only use text to express our opinions but there is also a growing number of opinions posted in different formats such as video.  A huge number of videos are posted every day whether YouTube, Facebook, and Snapchat. They record their opinions toward products using Cameras and upload them into the Internet which then helps other consumers to purchase a product. Video opinions provide multimodal data in terms of vocal and visual modality. That brings a significant need to build tools that combine text and video data that could build a better emotion and sentiment analysis model. In addition, Most of these tools whether the unimodal or the multimodal focused on the English language while some other languages such as Arabic which is the official language of 23 countries has a lack of researches that could improve the Arabic Sentiment Analysis tools and become easier and beneficial. In the last couple of years, there has been a growing interest with respect to Arabic sentiment analysis and more researchers are trying to fill the gap. However, to the best of our knowledge, there has been no research that focused on multimodal Arabic Sentient Analysis. So, in this research project, we will try to build state of the art Arabic sentiment analysis tools that combine textual, vocal and facial format.

Inflammatory bowel disease (IBD) is a common cause of chronic ill-health among young people in Scotland (prevalence estimated at 1 in 200 for adults and 1 in 2000 for children, with a peak incidence in the 2nd and 3rd decades of life). The major forms of IBD, namely Crohn’s disease (CD) and ulcerative colitis (UC), all too often confer a lifetime of the unpleasant, intrusive and potentially dangerous burden of intestinal inflammation on the individual. Typical symptoms include abdominal pain, diarrhea, weight loss and lethargy. These adversely affect schooling, work attainment, psycho-social well-being and sexual health. The health economic burden of IBD is considerable.

Existing treatment modalities are limited by lack of efficacy, unacceptable toxicity, and poor patient acceptability. Major surgical intervention is frequently required (>50 % in CD; 20% in UC), with a high risk of disease recurrence. Moreover, our ability to predict which individual will follow an aggressive disease course to enable targeted therapy is sorely lacking.

The proposed combination of off-line, on-line, supervised and unsupervised machine learning based prospective predictive analysis helped by addressing the following key questions: -

1. How can the prospective prediction of IBD disease flare be improved in high-dimensional real-world clinical settings?
2. How should the prospective methods be adapted to account for quantifiable differences between training and test datasets?
3. How should existing biological/genomic, microbial and clinical knowledge be exploited to improve predictions?
4. How can predictions of diseases and related complex traits be improved by stratifying patients into distinct groups/clusters based on heterogeneous genomic and clinical signature?
5. Which minimal sets of variables are needed to identify patient groups/clusters and make accurate, clinically actionable predictions within each group. 

Website Link : http://www.doctordotmalik.com/ ?page_id=90

Millions of users worldwide are sharing content using the Peer-to-Peer (P2P) client network. While such new innovations bring benefits, there are nevertheless some dangers associated with them. One of the main threats is P2P worms that can penetrate the network, even from a single node and then spread very swiftly. Many attempts have been made in this domain to model the worm propagation behaviour, and yet no single model currently exists that can realistically model the process. Most researchers have made use of disease epidemic models for modelling the worm propagation process. Such models are, however, based on strong assumptions that may not necessarily be valid in real-world scenarios. In this work, a novel analytical model is proposed, one that considers configuration diversity, infection time lag, user-behaviour and node mobility as the important parameters that affect the worm propagation process. The model is flexible and can represent a network where all nodes are mobile or a heterogeneous network, where some nodes are static and others are mobile. A complete derivation of each of the factors is provided in the analytical model, and the results are benchmarked against recently reported analytical models. A comparative analysis of simulation results shows that our proposed model, compared to the existing state-of-the-art analytical models, indeed represents a more realistic picture of the whole worm propagation process.

In this PhD research, we explore the novel use of Interactive Simulation to enhance the effectiveness of learning in higher education. We ran lab interventions using interactive simulation models, specifically agent-based simulation, to teach the concept of Spatially-explicit predator-prey interaction to undergraduate students in Ecology. We also developed an interactive simulation model (Agent-based Simulation) for a lab interventions with Aquaculture undergraduate and postgraduate students to teach them about sustainable fishing and the concepts of Total Allowable Catch and Maximum sustainable Yield and Catch Per Unit Effort using a simple biomass based production model. This is an exploratory study using mixed methods (Qualitative + Quantitative) tools.

Opinion mining and sentiment analyses offers a huge, but widely unrealised, potential to advertisers, companies and organisations and the public. Current sentiment analysis techniques are naïve as they are text based and only recognise certain keywords, which are far from the goal of true natural language understanding. The next level of fine-grained sentiment analysis associates emotions or attitudes with extracted text, and is of even greater commercial value. Prof. Hussain at Stirling has pioneered commercially-relevant, cognitively-inspired sentic computing techniques for concept-level social Big Data analysis, whose sustained international industrial impact was submitted as one of the School’s “REF2014 Impact Case Study”, and received an “Outstanding” evaluation.

In partnership with Industrial Supervisors, multi-modal fine grained sentiment and predictive-analytics will be researched, developed, tested and integrated in at least one existing commercial application (the company’s on-line wedding services management system) to provide a more natural, emotion-sensitive multi-modal human computer interaction. The core next-generation (multi-modal sentic) engine will be adaptable to other commercial real-world applications, including on-line travel management services – a global web-based travel search company based in Edinburgh, Skyscanner (UK) Ltd, has already expressed an interest in piloting the novel multi-modal analytics engine in their commercial Big Data system.