The Institute of Informatics and Telecommunications congratulates the five researchers who have been awarded their Doctorates of Philosophy (PhDs) in 2021.
Dr Eleni Litsa concluded her PhD thesis titled Elucidating Metabolism through Machine Learning.
During my PhD, I developed computational approaches based on Artificial Intelligence and Machine Learning in an effort to develop intelligent systems that can automatically extract information from metabolic datasets and develop tools that can assist metabolic studies. Metabolic reactions are chemical reactions that take place within an organism and play a crucial role in sustaining life. Such reactions break down the food we consume to produce energy and chemical matter to support our bodies. The same mechanisms process any chemicals that may end up in our bodies, either intentionally such as therapeutic drugs or unintentionally such as pesticides, to minimize potential harm. Over the years, there have been intense efforts to study and register information on metabolism of various organisms. Systematic analysis of these data can greatly advance development of therapeutics and diagnostic techniques.
Dr Nikiforos Pittaras concluded his thesis titled Beyond Deep Learning: Enriching Data Representations for Machine Learning Tasks.
My PhD focused at developing techniques that improve performance and explainability via the utilization of existing knowledge. To this end, I investigated Deep Learning methods that exploit curated, structured information from external resources, such as ontologies, knowledge bases and semantic graphs. I focused on the data representation component of a learning system, i.e. the mechanism of how real world objects are represented in the computation, with the study covering different problems (classification, summarisation and clustering) and data types (text, images, and audio). Novel representation systems were developed, that jointly considered data content, semantics and relationships between instances, mined from conceptual graphs and lexicons. The knowledge-augmented systems were shown to provide improved task performance and transparency, illustrating the potential of knowledge utilisation in learning tasks.
Dr Athanasios Davvetas concluded his thesis titled Evidence Transfer: A Versatile Deep Representation Learning Method for Information Fusion.
My PhD is about the investigation of an efficient, effective and robust deep learning method for intelligent information fusion. Deep learning enables the extraction of semantically rich, meaningful and dimensionally smaller data representations. Such representations are usually products of observing data from a single view. Introducing information from additional external sources can provide a multi-perspective view of the problem that may lead to increased effectiveness for the task at hand. The extraction of augmented representations, as a result of intelligent information fusion, should enable their generalisation and reusability, ensure robustness against noisy or malicious information sources and enable domain experts or policymakers to make decisions based on multi-view perspectives and heterogeneous data sources.
Dr Marios Pappas concluded his thesis titled Online assessment of Mathematics Difficulties for school-age children.
Mathematics Difficulties during early childhood are mainly found in arithmetic skills and problem solving. These difficulties are related to working memory deficits, as well as attention deficits. The purpose of the PhD was the development of an open access online screening tool, which will be able to identify students at risk of mathematics difficulties in a timely and valid manner. The BrainMath scale comes to cover the shortage of screening tools for Mathematics Difficulties, assessing mathematical and cognitive skills from the first grades of primary school, without being time consuming, utilising new technologies. Findings of the pilot study show that the proposed scale is a valid and reliable screening tool which can be used for educational or research purposes.
Dr Elias Alevizos concluded his thesis titled Complex Event Forecasting.
As analytics moves towards a model of proactive computing, the requirement for forecasting acquires more importance. Systems with forecasting capabilities can play a significant role in assisting users to make smart decisions as soon as critical situations are detected. Being able to forecast that certain patterns in a stream have a high probability of being detected, before they are actually detected, can help an analyst focus early on what is important and possibly take a proactive action. The need for event forecasting as a means for proactive behaviour has led to proposals about how forecasting could be conceptualised and integrated within a complex event processing system. However, such proposals still remain largely at a conceptual level, without providing concrete algorithms. The goal of this work was to provide a theoretical basis and build a prototype system for forecasting the occurrence of complex event patterns. This was achieved by advancing the state-of-the-art in automaton models, going beyond classical automata, so that patterns can have the expressive power required by complex event processing applications. A probabilistic framework was subsequently be used, so that the behaviour of these automaton models may be quantified in a way that allows for producing forecasts with confidence.