MBIE-funded University research could open up new technology
Never mind painstakingly filling out forms to identify scratches when you hire a rental car, imagine a split second scan which could analyse damage before and after use.
Or a device that with one quick scan can tell you whether that dress on the rack is a match for your current wardrobe.
A University of Waikato team has received funding for a research project which could make it far easier for a huge range of everyday technological solutions to get off the ground.
A $1 million Ministry of Business, Innovation & Employment (MBIE) Endeavour Fund grant has been announced for a three-year User Friendly Deep Learning research project led by computer science professors Geoff Holmes and Eibe Frank.
The research could mean that for the first time, complex machine-learning technology could be made available in a simple form for a huge range of commercial applications.
Machine learning, which is linked to Artificial Intelligence, is the study of algorithms and statistical models that computer systems use to perform tasks without instructions, by relying on patterns and images.
Many economic, environmental, and social questions require accurate decision-making based on data consisting of a large number of highly correlated variables, says Prof Holmes.
“Examples include orchardists wishing to estimate crop yield, fertiliser companies wishing to measure the quality of their products, paediatric neurologists wishing to identify cognitive disabilities in children from an analysis of their physical movement, and conservationists wishing to estimate species abundance in streams and rivers.”
Most organisations face significant barriers to the use of this technology because of the set-up cost involved and the lack of expertise in New Zealand. Companies such as Google and Microsoft provide off-the-shelf solutions for standard tasks such as object identification and image classification, but it is still often very challenging to obtain good tailor-made machine learning solutions for specific problems such as those above, particularly when little data is available for machine learning.
“We believe that we can develop a user-friendly software platform that removes this bottleneck,” says Professor Holmes.
“This platform will unlock access to deep learning technology for a much wider sector of the New Zealand economy.”
He believes access to the technology could provide a huge boost for small and medium enterprises that can’t invest in the people needed to do the required technological development.
“The idea is that we’re trying to construct a system so that anyone with an idea or a need can have a prototype solution built that validates their proposition.”
Another unique facet of the research proposal is the ability to tweak models so they reach the ultimate in accuracy, via direct user involvement with the system.
“So what we’re doing differently here is putting the human in the modelling loop as well,” says Professor Holmes.
“Normally what would happen is a human would produce a dataset at the start for the computer to use and then be pretty hands off. That would be the extent of their involvement in the process and their only opportunity to input knowledge into the process.
“This model has the
user much more intimately involved in the cycle of doing the
The research project officially begins on October 1.
Professors Holmes and Frank will be assisted by Professor Mark Apperley, Dr Te Taka Keegan, Peter Reutemann and Dale Fletcher – all of whom bring their specialist skills to the subject.
The team’s first task will involve building a number of prototypes early on to test out responses.
“We will try to construct the whole system in a sort of vanilla form quite quickly so that we can get some user interaction going. It’s good to do that testing early on rather than leaving it to the end.”
He is excited and confident about the outcomes of the research.
“The Endeavour Fund is about addressing a problem
that’s not been solved. And I've always believed in this
idea which I think has both practical and commercial