AI@MS Journal Club

The Journal Club is currently having a pause, as it lost its momentum during the pandemic. We hope to soon restart this activity again.

The format for the journal club is very open ended. Each speaker chooses a paper or a topic of choice that he or she is interested in. It does not have to be a very recent result, but could just was well correspond to some basic part of, say, deep learning, or some open problem that you are thinking about and wish to discuss.

We meat roughly once per month, and every meeting has 1-2 speakers, of about 25 minutes each, including questions/open discussion. The speakers create new posts in Yammer for their talk, including links to relevant papers or other literature. Slides can also be uploaded to the same post after the seminar.

If you want to participate in the journal club please contact Daniel Persson ( at the Department of Mathematical Sciences.

Past seminars


Seminar 1: Filip Wikman (Department of Mathematical Sciences)
Title: “Deep learning as an optimal control problem”

Seminar 2: Rebecka Jörnsten
Title: "Wide neural networks of any depth evolve as linear models under gradient descent"


Anton Johansson (Department of Mathematical Sciences)
Title: “Bayesian Neural Networks”

Olof Zetterqvist (Department of Mathematical Sciences)
Title: “Finding outliers using the influence function”

Annika Lang (Department of Mathematical Sciences)
Title: “Deepest learning using stochastic partial differential equations”
Literature:  (preprint),  (journal ref)


Seminar 1: Jimmy Aronsson (Department of Mathematical Sciences)
Title: “Restricted Boltzmann Machines”
Seminar 2: Fredrik Hellström (Department of Electrical Engineering)
Title: "Neural Loss Surfaces and Local Minima”

Seminar 1: Petter Mostad (Department of Mathematical Sciences)
Title: "AdaNet: Adaptive Structural Learning”
Seminar 2: Devdatt Dubhashi (Department of Computer Science and Engineering)
Title: "Optimal Transport and Domain Adaptation”

Seminar 1: Giuseppe Durisi (Department of Electrical Engineering)
Title: "Opening the Black Box of Deep Neural Networks via Information”
Seminar 2: Christoffer Petersson (Deep Learning Team, Zenuity)
Title: "Measuring the Intrinsic Dimension of Objective Landscapes”

Seminar 1: Daniel Persson (Department of Mathematical Sciences)
Title: "Deep Learning and Renormalization”
Seminar 2: Johan Jonasson (Department of Mathematical Sciences)
Title: “Variational Inference"

Page manager Published: Thu 03 Mar 2022.