Deep learning language

May 14, 2018

Developer and data scientist Miles Thompson recently started working with Dragonfly on the Te Hiku Media Kōrero Māori project – applying machine learning to transcribing spoken Māori.

Miles has worked off and on in machine learning throughout his career and is now relishing the opportunity to focus on it in a more full time capacity.

While in New York, he worked with big data and C# programming in his main job but also did some other data science and python work, including machine learning.

“Recently I was involved in creating a theatre production that used a degree of machine learning to create a different performance every night. In The Great Outdoors, we harvested recent comments on the internet and using machine learning, we found and arranged comments that were then spoken in the performance. It was fascinating to see my programming work presented in that setting.”

Miles’ day job for five years was chief technology officer for the financial research provider CreditSights. After two children came along, he and his partner moved back to New Zealand and Miles carried on working for the company as a technical architect.

During those years he was involved in projects with Edward, including a venture that incorporated machine learning and AI. Although he’d studied neural networks at university, he had only used them in a research context prior to that time.

“It was a powerful approach but needed a few more years for hardware to catch up with the algorithms. At that stage people had almost given up on machine learning and neural networks because it was so difficult to get them to work.”

Then in 2014 he went along to a talk at Dragonfly (at the invitation of his brother Finlay) given by Marcus Frean from Victoria University who presented the latest techniques coming out of the University of Toronto. The term ‘deep learning’ was coined around that time, to recognise this new, deeply layered approach.

“It’s a bit like machine learning 2.0. We’re applying some of these new techniques to the Kōrero Māori project, and iteratively improving the accuracy of the model we’ve developed so far. Dragonfly are doing cutting edge work – it’s a very exciting place to be. Things are changing rapidly in the area of machine learning at the moment and we’re at the forefront of that.”

Dragonfly Data Science is helping on Kōrero Māori, a project building speech recognition software for te reo Māori. The project is led by Te Hiku Media, and supported by Te Pūnaha Matatini. Listen to a story on RNZ about the project.

Miles is available for other data and machine learning projects and can be contacted via Goodbright Consulting.