Dr Enda Barrett, NUI Galway. Image: Laura Barrett
Dr Enda Barrett of NUI Galway is training AI to quickly detect when someone may be drowning and alert authorities.
After completing his undergraduate degree in IT, Dr Enda Barrett joined the Enterprise Computing Research Group in NUI Galway. This was followed by a master’s in science, with research in classifying arrhythmias for ambulatory patients.
In 2013, Barrett completed a PhD in computer science and, after a two-and-a-half-year stint in industry, he returned to NUI Galway as a lecturer at its school of computer science. To date, he has published more than 40 research articles including 12 journal papers, four patents and 25 conference/workshop papers.
‘Training vision systems capable of making accurate detections in the widest possible number of scenarios is a data-hungry task’
– DR ENDA BARRETT
What inspired you to become a researcher?
As a young boy, I was always very interested in science and excelled in those subjects at school. Chemistry was one of my favourite subjects at second level and I particularly enjoyed the experimental side of it.
I can still remember mixing hydrogen peroxide with potassium iodide and washing-up liquid to produce an ‘iodine snake’, which quickly turned into a foam party among the class.
Can you tell us about the research you’re currently working on?
I’m currently working on a Science Foundation Ireland-funded project called ALIVE (Autonomous Lifeguard and Vision Environment). The purpose of the project is to accurately detect and monitor individuals in aquatic environments such as oceans, beaches, lakes and pools. This aims to prevent drownings by alerting authorities as quickly as possible when individuals stray into hazardous areas or fall into the water.
Using a computer game engine, we model a person under a variety of dangerous test conditions which would be difficult to get data on in real life. We then use this data to train more accurate domain-specific machine learning classifiers, which have a greater accuracy than those trained with stock images of people.
This approach also allows us to cater for scenarios where only a partial view of an individual is available, such as if the person is partially obscured by rocks or waves.
In your opinion, why is your research important?
Most major cities have large bodies of water flowing through them. Galway, in particular with the Corrib river, has the fastest-flowing city river in Europe.
Unfortunately, there are a number of fatalities each year as a result of people falling into the water accidentally. If the research investigated as part of the ALIVE project could be successful in saving even a single life, then I feel it will have been worth it.
What commercial applications do you foresee for your research?
In the ALIVE project, we see commercial opportunities in providing a low-cost alternative to search helicopters.
Using off-the-shelf drones, the vision system we have developed can quickly scan a larger area for bodies in the water and aid in faster detection of people. We also see this as an aid to water safety personnel, enabling them to better manage the tracking and monitoring of individuals in the water.
What are some of the biggest challenges you face as a researcher in your field?
Training vision systems capable of making accurate detections in the widest possible number of scenarios is a data-hungry task. While a well-designed classifier will generalise to unseen environments, the more data you have describing the particular scenario of interest, the better your classifier will perform. Therefore, it’s always challenging getting high quality data from the widest possible number of scenarios.
Are there any common misconceptions about this area of research?
A huge amount of blood, sweat and tears goes into model training in machine learning domains. However, whilst there is a significant design effort involved in getting a learner to an acceptable level of performance, the trained model can often act as black box, leaving the designers scratching their heads as to why certain decisions were made.
If we can better explain how the generated models arrive at their decisions, we can improve the algorithms further. A large subfield within AI, known as ‘explainable AI’, is focused at addressing this issue and is gaining a lot of traction lately.
What are some of the areas of research you’d like to see tackled in the years ahead?
Machine learning and AI in general have experienced a boom of late and are seen as expert solutions to a wide variety of problems.
I’d particularly like to see more focus on areas which can result in social good and I have a particular interest in the work being done by the University of Southern California Center for Artificial Intelligence in Society.
Are you a researcher with an interesting project to share? Let us know by emailing editorial@siliconrepublic.com with the subject line ‘Science Uncovered’.
