The aim of this project is to investigate the influence of environmental and observational effects on Harbour Porpoise survey counts from the NPWS monitoring programme conducted across three spatial areas of conservation in 2007, 2008, 2013-2016 and 2018. First, the detection probability function will be estimated using distant sampling methods extended with covariates in order to understand their effect on the function. Then, the probability of detection will be incorporated into spatial model (density surface model). Finally, using the estimated density surface model, the power to detect changes at given sampling intensities (survey intervals, number of surveys) will be calculated on a site-by-site basis. The project team consist of 2 modellers and 3 marine mammal experts.
Read MoreThis project is a PhD Cullen MI award. It is a multidisciplinary project that aims to bridge the gap between marine sciences and machine learning in detecting climate change related anomalous events. Anomalies are data points or patterns in data that are unusual and do not conform to a notion of normal behaviour. Anomaly detection is the task of finding those patterns in large volumes of data. It is a challenging topic because anomalous events can be abrupt, especially when they are related to climate change, to which marine ecosystems can have a complex response. Automatically detecting and correctly classifying anomalies is also difficult because data is multidimensional. The overall goal of the project is to provide a deep learning pipeline for online short-term prediction of anomalous and/or extreme marine events which may impact the commercial aquaculture sector.
Read MoreThis project has the cross cutting disciplinary nature and aims to develop and apply a deep learning algorithm to assist in predicting the clinical outcomes for treated stroke cases. Reason: Randomized control trials (RCT) can be limited when assessing the next generation of medical devices due to the length of time and associated costs. The introduction of machine learning can improve the surveillance of medical devices and lead to profound changes in the practice of running clinical trials. Medically acquired data is inherently unbalanced and require deep learning networks to address this. Therefore, there is a need for virtual studies using more advanced computational algorithms. Methods: 200 acute stroke treated cases from various medical centers in Europe will be made available through international collaboration. This data consists of a simple CT scan with a resolution 512 x 512 x 40 data points (>10 Mln) and size of 1-4 GB per case. This data will be mapped to new representations and used to make predictions using deep learning algorithms.
Read More