Machine learning for computers takes a major step forward
Thomas Schön, professor of Automatic Control at the Department of Information Technology has received a framework grant of 29 million SEK from SSF – the Swedish Foundation for Strategic Research for the project Automating System SpEcific Model-Based Learning (ASSEMBLE). The project group includes David Black-Schaffer at the IT Department of Uppsala University, researchers at the Royal Institute of Technology, Stockholm, and a number of companies.
What is the project about?
“The project is to do with machine learning, that is how computers can learn to reason and react based on measured data. Machine learning is a fast growing field on the borderline between computer science and statistics. The main innovation of our new project is to create a kind of bridge between advanced algorithms and users so that the algorithms we construct will be come to good effect more quickly. This is why we are going to compose a new, specially written programming language to differentiate models and algorithms in a way which has previously not been possible in machine learning.”
Who are the intended users of the new programming language?
“Often it is engineers of various kinds. But we also work with companies such as Autoliv who produce automobile safety systems. In their case, we collect data from different cameras and sensors around and inside the vehicle which is translated into usable knowledge which enables the vehicle to make decisions itself and also provides information to the driver. This knowledge we collect together in models which are then used to make decisions such as braking, accelerating and turning.”
“Another application we are working on for Karolinska Institutet is so-called ‘cell tracking’, which is used to automate the study of cells for biomedical analyses.”
“But the task before us is much more general than these. Viewed in a larger perspective, I see our project as an important contribution towards advancing Sweden’s standing in the field of machine learning. This is why we also have industrial partners to ensure that the project really has some practical benefit.”
What is the project’s greatest challenge?
“The challenge lies in describing not only what we know but also things that are not so certain. As luck would have it, we have good mathematician friends mostly specialising in probability theory and statistics and we have worked a lot on how to describe uncertainty. In my field, machine learning, much has also happened in recent years which means that we can make models on a much larger scale.”
What will the 29 million SEK be spent on?
“The money will be spent on realising the ideas we have. And so we will continue to put together an interdisciplinary team and add to it further doctoral students and post-docs. The main thing is to get in more of the right people to drive the project forwards.”
“We also have a new first-cycle course in statistical machine learning in the works. Courses in this field provide extremely important structural capital for research but also for industry and the community at large.”
David Black Schaffer: Delivering efficiency with heterogeneous computation
23 June 2016