I’ve been with UNSILO for just one and a half weeks. Before working here, I was studying software engineering at Aalborg University – about 100km north of Aarhus. The software engineering course is similar to computer science, sharing most of the modules with data science, but with a more practical emphasis. As part of the course I worked on ten software projects.
Which of the projects did you most enjoy?
The last, and biggest project, which was a machine-learning application. It was with a company based in Aarhus – my Aalborg professor is the CTO of this company. The project was about learning representations of patient data, using unsupervised learning.
How did you find out about UNSILO?
I went to an event in Aarhus where Mads Rydahl, the co-founder, was giving a talk.
How do you find working in a startup?
Much better than the multinational I was working –it was the Danish office of a multinational US corporation. I had a part-time job there while I was student, and it gave me some experience of a large corporation. But I’ve also worked for very small companies – I gained some experience working with a small entertainment company in Copenhagen, creating animations and apps for mobiles. I was actually born and grew up in Copenhagen.
So you have experience of the two biggest Danish cities! Which do you prefer, Copenhagen or Aarhus?
Actually, I prefer Aarhus -because I always lived in the suburbs of Copenhagen, while here in Aarhus I live closer to the centre and I can come to work by bike. Aarhus is compact and accessible and close to the countryside. It has a denser layout than Aalborg, which means you can be in the country sooner.
How did you get into machine learning?
We had to do a pre-specialisation in our last year, and I did some unsupervised learning using Recurrent Neural Networks. This project involved taking patient data, and creating a vector, capturing what happened at five-minute intervals For example, the patient was admitted, then was seen by a doctor, then received treatment. All of this can be put into a sequence model, and in this way you can create an embedding of the entire patient history. After you have done this for many patients, you can then compare similar sequences – what in Danish we call forløb – and draw conclusions from that comparison. By comparing records that have similar sequences, it is possible to make an educated guess about what the problem might be. For example, if three patient records follow a similar sequence of events, there is a reasonable likelihood that a fourth patient with the same initial event sequence may have the same condition.
And was your project successful?
Well, we got better than random results! We think the model was good, but we didn’t have enough computing power to build a full model. It certainly gave me a clear insight into the possibilities of machine learning, including UNSILO’s speciality, which is applying machine learning to text.
What projects are you involved with at UNSILO?
Currently I am doing some work with Tensor Flow, integrating some part-of-speech tools.
If you weren’t working with UNSILO, what do think you might be doing?
I’ve always been interested in biology, or more specifically bioinformatics, such as sequencing of DNA. My brother is currently studying biotechnology in Copenhagen.
Many thanks, Martin.