I am a neuroscientist & statistician and use machine learning and data analytics to link cell types, circuits and computations in the healthy and diseased visual system with a particular focus on the retina. I work in the labs of Matthias Bethge and Thomas Euler at the University of Tübingen and collaborate closely with Andreas Tolias at Baylor College of Medicine. See my Research page for my current projects and interests.
Christian Behrens, a new PhD student, has joined our lab to work with me on data analysis and modeling in the retina. He studied theoretical physics in Munich and Heidelberg before coming to Tübingen. He will join the team working on the DFG grant we have been recently awarded, collaborating with Tom Baden, Katrin Franke and Thomas Euler.
We have recently started to implement PyCircStat, a python package for statistical analysis of circular or angular data. It builds on the matlab toolbox for circular statistics I have developed a few years ago. The python package is still in the early stages of development, but a number of features are already working:
You are welcome to check out the toolbox and try it – let us know if you find bugs, or have feature requests. Thanks!
On September 2nd, I will give a talk at the workshop “Characterizing Natural Scenes: Retinal Coding and Statistical Theory” organized by Arno Onken and Jian Liu about our recent Nature Neuroscience paper “Population code in mouse V1 facilitates readout of natural scenes through increased sparseness”. I am looking forward to an exciting workshop program and interesting discussions!
My grant application “Emergence of retinal ganglion cell response diversity from synaptic interactions in the inner retina – a combined approach of two-photon population imaging and computational modeling” together with Tom Baden has been funded by the DFG. We are excited to get started with the work!
Two of our papers came out during the last few weeks. In “Population code in mouse V1 facilitates read-out of natural scenes through increased sparseness” (published in Nature Neuroscience), we show that higher order correlations in natural scenes lead to increased population sparseness, improving the representation of natural scenes in the population code. In “State dependence of noise correlations in macaque V1” (published in Neuron), we show that global activity fluctuations in neural activity lead to noise correlations during anesthesia, and that removing these global fluctuations recovers the awake correlation structure.
A new review on population coding by Maoz Shamir is appearing in the April issue of Current Opinion in Neurobiology. Shamir discusses how correlated noise negatively impacts neural coding, when the noise overlaps with the signal, and how this effect can be overcome by introducing heterogeneity into the population model. Importantly, he points out that current population coding models are heavily underconstraint, such that additional assumptions are needed for testing them. The review references much of our work on population coding over the past 5 years or so, which was nice to see. Overall, this very interesting special issue on theoretical and computational neuroscience assembled by Adrienne Fairhall and Haim Sompolinsky is highly recommended
Last week, I got the Klaus Tschira Award for the Public Understanding of Science for my dissertation and an article describing my thesis work in an accessible fashion. The article appears in Bild der Wissenschaft, a German popular science magazine and can be found here: ‘The programming language of the brain‘ (in German).
When we originally submitted our 2008 paper on the orientation selectivity of LFP gamma band activity, the response of one of the reviewers was: “It simply reeks of artefact.” We had found that the LFP gamma band power across our whole array of tetrodes spanning ~1 mm was tuned to the same orientation (give and take) and the preferred orientations hardly correlated with that of the multi unit activity. But after going through many rounds of analysis, I was convinced that our finding was correct and we eventually published the paper.
Nevertheless, I was quite relieved when another group independently obtained the same result: Jia, Smith and Kohn reported in their 2011 paper that for large stimuli (as also used in our study), all LFP sites in an array spanning ~4 mm show the same preferred orientation. In addition, they verified this finding independently with single electrodes and still obtained the same results. Interestingly, they found that when using smaller stimuli, LFP sites exhibit different orientation preferences which correlate well with the multi-unit orientation preference at the site.