At #DataMash, We spoke to Senior Research Scientist Isabel Sargent from the Change and Business Innovation team at Ordnance Survey about Machine Learning. This is what she had to say.
(Transcript below)
I’m Isabel Sargent – I work in Business Change and Innovation in Ordnance Survey. We’re the team looking at how we capture data; what platforms we use; sensors; – and then how we process those data to produce what our customers want; what our internal processes need.
So here you’ve got some slides on Machine Learning. How important is that in the services you’re offering to customers?
We feel that machine learning is going to offer us a great deal in terms of trying to interpret data quickly and flexibly, so that we can actually respond more quickly to what customers want.
It seems to be quite a tremendous buzz word at the moment…
Oh massively, yes…
About machine learning…
– it’s ridiculous!
…because of ‘big…
And Worse…worse…
Yeah?
…we’re using deep learning; deep networks…
Right…
…which is massively hyped at the moment.
err…now…what…how might you allay any fears that people might have that anything you’re doing might encroach on personal data? Is that something that you try to steer clear of?
We’re not touching personal data at the moment at all, and have no particular intentions to. We’re solely focusing on data that’s been sensed by – in this case – an aerial camera, standard things that we use. This is actually an imagery layer image – simply three-bands – we could go beyond it; we could look at four bands, we could look at other sense data, but there’s nothing personal in that.
What kinds of things are people interested in finding out using machine learning? Is it simple things like: where’s the boundary between land and water, or is it more complex things?
Well: you could potentially use it for any of those questions you might have. I think it’s particularly interesting where your problem is more of a subjective nature. Where it’s something where you can’t write a rule to discriminate, but people know…they know what the difference is between one thing and the other, but writing those rules is difficult, so machine learning helps you to maybe learn those rules. The way that we’re using it in this project is to not pinpoint anything in particular; we’re not trying to solve a particular customer’s problem. What we’re looking at is can we preprocess all the data that come in, in a way that then allows us to rapidly respond to questions as they come. So it’s the first stage; it’s going to be slow; we’ve got to do a lot of discovery; finding out what’s in the data – has it got enough in the data? – and then at some point we’re hoping to just go with it, run the processing online as the data come in, and then when customers come to us with questions, we’re hoping it should preprocess and then we can rapidly answer those.
It’s really about anticipating what the users want, and then getting to a certain stage first…yeah.
Absolutely.
That’s really interesting so thanks for your time and for telling us about machine learning.
