Jim O'Donoghue - Transfer Talk - 25th August 2014

Video Category: 
Transfer Talk

Title: Modelling High-Dimensional, Sparse and Sequential Data with Deep Networks

Supervisor: Dr. Mark Roantree


Today's computer systems accumulate data at almost unimaginable rates with many applications generating large silos of data. Online applications vary from banking, shopping, travel and even to social interaction as we conduct more of our requirements online. In areas such as healthcare, there is now a deliberate attempt to capture, record and analyse people's behaviours and lifestyle in order to predict or possibly prevent future health issues. The In-Mindd European project which funds this research is focused on the area of dementia and one of its functions is to use a longitudinal study - MAAS (Maastricht Ageing Study) - to determine the likelihood of developing dementia, based on health and lifestyle. Data mining is an important process in providing accurate predictions but the characteristics of many datasets, including MAAS, make prediction using normal techniques unreliable and often intractable. In this research, we examine a complex method of data mining, called deep learning, to determine if higher degrees of accuracy can be achieved. While traditionally used in the areas of speech and vision, we are attempting to apply these methods to more traditional data in order to optimise accuracy and overcome certain data problems with an evaluation across different domains and datasets to test the robustness of our approach.