– This all happened very fast. One and a half years ago many of the current activities did not exist. It is a new area for many of us and it feels really exciting.
The start of the research area AI@MS was the Wallenberg initiative WASP (Wallenberg AI, Autonomous Systems and Software Program). Within this, a special call was made about the mathematics behind AI with, among other things, 18 PhD studentships. A meeting was held at Mathematical Sciences to produce proposals for doctoral projects, and the result was successful for Chalmers who got six projects, four of them at Mathematical Sciences. Daniel Persson supervises one of the four projects, the other supervisors are Annika Lang, Petter Mostad, and Johan Jonasson/Rebecka Jörnsten. The PhD students who began this autumn are Jimmy Aronsson, Filip Wikman, Anton Johansson and Olof Zetterqvist.
– I think it is far-sighted of the Wallenberg Foundation to focus on the underlying mathematical structures of AI. AI is already used by the industry, but it is not always understood why some things work and others do not. All of us at Mathematical Sciences who has projects related to AI have different perspectives and entry points, and we have not interacted much before, so now we create new networks.
AI@MS Journal Club
One such new network is the AI@MS Journal Club which also includes those who work with Chalmers two other WASP project in AI/Math, at the departments Electrical Engineering and Computer Science and Engineering. In the many seminar series that the department hosts the speakers often talk about their newly written articles, but here everything is so new. Instead, other people’s articles or work are discussed, to get a discussion started and to get to know each other. The meetings have been held once a month since September and despite the fact that they have not been announced the rumour has spread and the number of participants has grown. The lack of advertising has been intentional, with the idea that also a newly admitted PhD student should feel comfortable talking. Next term there will probably start an open seminar series that also has invited speakers, alongside the AI@MS Journal Club.
Daniel’s project is called Quantum Deep Learning and Renormalization and is about unsupervised (deep) machine learning where you try to create algorithms so the machine can train for example to recognise motifs in images on its own, without having to use annotated data in the training process. The purpose is that the machine should be able to recognise patterns and structures and be able to form new ones. This research is still in its infancy and Daniel uses ideas from theoretical physics, more specifically statistical mechanics and quantum mechanics, combined with information theory.
– There is a surge of interest for these projects. A person at Zenuity, who works with self-driving cars, is industrial supervisor for my project and also Volvo has been in contact. I held a Master’s course in differential geometry and several students came and asked about the AI projects so now I have three master thesis students working on various aspects of deep learning. It feels good to be able to attract interested mathematics students instead of sending them on to other departments.
Many mathematical AI projects
AI@MS is not only about the WASP projects. On the website you can read about eight other projects at Mathematical Sciences that connect to AI in different ways, they are, for example, about neuronal networks, point processes, artificial life, partial differential equations, image processing, medical diagnoses, and AI as a threat or opportunity. The projects cut across the department regardless of mathematical alignments and much work is done over the subject boundaries.
A parallel venture with WASP is CHAIR, Chalmers AI Research Centre, which had kick-off on March 4 and is for the most part financed by the Chalmers Foundation. They also have a call for PhD studentships, which certainly researchers at Mathematical Sciences will apply to although this initiative is not specifically aimed at mathematics like WASP. Within the WASP initiative, recruitment of an assistant professor is in progress with trial lectures held in March. The idea is to bring in a more senior person with previous experience, and the employment comes with a PhD student and four postdoctoral years.
– Yes, there is a lot going on and many people are pulling. I am already invited to speak at two conferences, otherwise the routine is that you have to publish articles first. Now we must ensure that we also have time to do research!
Photo: Klas Modin
Text and photo: Setta Aspström