Machine learning has many applications in the real world, from medical diagnoses to detection of contaminants in soil samples. And now, researchers are investigating whether it can be used to protect New Zealand’s biosecurity.
Pest species are a particular threat to our unique biodiversity, but how do we identify what is a pest and what is a harmless member of the local fauna and flora? University of Waikato Professor of Computer Science Eibe Frank, together with Dr Varvara Vetrova at Canterbury University and Associate Professor Michael Cree at Waikato, is working on a program to help everyday Kiwis identify pest species using their smartphone.
Professor Frank will discuss some of the underlying machine learning technology at an upcoming public lecture on 16 July, drawing connections to work on statistical species identification that pre-dates the computer age. He will address how we can use machine learning to identify species automatically, by learning from photos that have been labelled by experts.
Professor Frank’s research focus is machine learning, data mining and artificial intelligence (AI) and their applications for the real world. As a PhD student, he was instrumental in the development of WEKA, the University’s award-winning open-source machine-learning software platform, and he continues to support its development and international use.
“It turns out learning to discriminate species from data is one of the oldest, if not the oldest, application of machine learning,” says Professor Frank. Ronald Fisher, a statistician and biologist, described a method for linear discriminant analysis in 1936 and applied it to the classification of species of iris flowers. His method is now a classic technique for supervised machine learning – learning from expert-labelled observations – but it was published years before computer scientist Alan Turing discussed the idea of learning machines in his seminal 1950 paper on ‘Computing Machinery and Intelligence’.
Professor Frank says until recently, expert knowledge was required to define features that can be input into machine-learning models to establish a representation of the problem that makes it amenable to machine learning. “However, recent developments in the field of artificial neural networks, sometimes referred to as ‘deep learning’, have changed this. Artificial neural networks can automatically learn a set of features to represent images, often yielding more accurate image classifiers than those based on ‘hand-crafted’ image features. This opens up new opportunities for automatic species identification based on photos taken with digital cameras.”
Professor Frank says AI can often be overhyped or oversold, but insists there are many applications where machine learning can improve outcomes for people and increase productivity for organisations. “Currently, New Zealand is lacking university graduates with significant knowledge in this area, but my colleagues in the Waikato Machine Learning Group and I are working hard to change this.”
The lecture, ‘Learning to discriminate species from data: Then and now’, is on Tuesday 16 July at 5.45pm in the Gallagher Academy of Performing Arts, with refreshments served from 5.15pm. The lecture is part of the University’s Hamilton Public Lecture Series and is free and open to the public. Register your attendance.