My name is David Kofoed Wind and I am a PhD-student in the Cognitive Systems Section at DTU Compute. My PhD-project relates around machine learning and data science, covering a broad set of topics. Currently I am working on three different projects: Analysis of data from the , peer grading of students in university courses and machine learning on graphs.
I am part of the Sensible DTU Project, an attempt to analyse the behaviour of a large population of university students over multiple years using smartphones. We have distributed 1000 smartphones over multiple years to students at The Technical University of Denmark, and are tracking as much as possible about the students: their location, wifi, bluetooth, calls, texts, academic grades, Facebook walls and much more. My work within the project has primarily been about using WiFi data for tracking location instead of GPS.
On the side of my academic research, I am co-founder of the startup www.peergrade.io
which is a service for providing peer grading-as-a-service to university courses. We have developed the entire software-system and it is currently in use at 4 courses at DTU and University of Copenhagen. As part of this project, we are looking into how we can use machine learning to make more accurate grading and to help students get better feedback on their reports.
My final project at the moment is about applying machine learning to graphs. In recent years, the use of graph databases (such as Neo4j) is getting more common. Storing data in a graph form is very intuitive and is often the best fit for a problem. Unfortunately, the process of applying machine learning methods to graph databases is not very well-developed, and state of the art methods still include hand-engineering of features and manual selection of weights. Together with some students, I am trying to rethink the way to extract features from graphs in order to do machine learning properly.
Besides my research and the peer-grading project, I am teaching a large course (130 students) titled “Computational Tools for Big Data” (www.toolsforbigdata.com
). The course focuses on the practical aspects of working with large datasets. A lot of students take years of theoretical courses on algorithms and software development – but still lack fundamental skills when it comes to using cluster computers or compiling the tools they need. In Computational Tools for Big Data we try to show the students all the hard problems that reality contains ( which is not always a pleasant experience).
I have been fortunate to be doing a PhD at DTU with professors Ole Winther and Sune Lehmann as supervisors, since this gives me an extreme freedom when it comes to choosing projects, teaching my own course and even running a business on the side.