About Me

Here, Steffen developed a method for producing identifiable attribution maps in time-series data using regularized contrastive learning. I applied it to synthetic and real-world grid cell data.

In this study, Cyril built a self-supervised pipeline for segmenting cells in 3D fluorescence microscopy volumes, reducing the need for manual annotations.

Lots of ANNs have shown grid-cell like activity in their hidden units. We ask if this is an artifact of the objective function used. Includes cute animals.

Here we were trying to build a neural network that simulates the egocentric to allocentric transformation circuit from visual areas to hippocampus. It works and we now also have a benchmark for ANNs that closely follows the four mountains test.

After lots of discussions in our lab meetings about the problems with eye-trackers we decided to build a deep learning framework that decodes eye position directly from MRI data.

DeepInsight – nicknamed the automatic John O' Keefe – decodes behavioral variables from wide-band neural recordings. This gives more objective measures of information content, compared to spike-based methods.

In this study, Matthis showed how directional tuning in the human visual-navigation network shifts depending on behavioral context during spatial navigation.

We saw this challenge and thought it would be nice to compete. Top 5 in the end but with one of the most efficient implementations and a autoencoder regularization.

Here we tested a setup which I introduced during my master thesis, to see if stimulating individual neurons can shift the activity of head direction cells in the presubiculum. It can!