Project Update: Dynamic Data Analytics through automatically Constructed Machine Learning Pipelines

Project update (January 2020) – Q&A with professor Bäck

When did the project officially start?

21 months ago (link to project)

How many people are on the project?

We have 3 PhD students at the moment and 7 supervisors are involved, 4 from the universities (Leiden, Eindhoven, Delft) and 3 from the companies involved with the project.

How did you start?

We started with a kickoff meeting to align the research and expectations of everyone involved. After that we looked into the state of the art technology and decided how to tackle the scientific research and the application tasks.

What was the biggest challenge thus far?

The biggest challenge at the moment is finding more application data in the energy consumption domain. In parallel, we are developing new automatic machine learning pipelines.

Is there a lot/enough contact with the private (paying) partners and how does the colabaration take shape?

We work with multiple private partners, with Honda Research Institute Europe being most tightly involved in the project. Honda has their own Academic research center and one of our PHD students spends three to four months per year on their location. The other private partners are involved to a smaller extent. They are learning from the project, so they gain knowledge through updates, and at the same time they bring in their experience in the application domain as well as their scientific background.

Did you changes the research questions during the research?

No, it is still the same question. We know what we had to do, only in the details there are some open questions that changed along the way. But the big two questions are still the same.

In which fase is the project at the moment?

At the moment, we have first results in developing automatic machine learning pipelines for multivariate time series forecasting and multivariate time series classification, and their application to the project’s application domains. These are energy forecasting based on energy consumption data and Parkinson’s disease diagnostics in video and EEG data. First results have been published.

What is the end goal?

Predicting the development of Parkinson’s disease on patients and also predicting the energy consumption of buildings to better control them are the application goals, while the fundamental research goal is to develop methods for automatic machine learning pipelines for time series classification and forecasting.

14 February 2020


Sign up for our newsletter and stay informed of the latest news Commit2data