Dominic Gonschorek

I am a Ph.D. student at the University of Tübingen

Here, I work in the lab of Thomas Euler, where I study the effects of neuromodulation on visual signal processing in the retina. In my work, I combine two-photon imaging, computational neuroscience and machine learning. As part of my Ph.D. program, I worked in the lab of Olivier Marre at the Institut de la Vision, Paris. Prior to that, I did my masters in Neuroscience at the University of Oldenburg, studying sensory and computational neuroscience. During that time, I have been a fellow of the SmartStart program; a joint training program in computational neuroscience launched by the Bernstein Network Computational Neuroscience and the VolkswagenFoundation. I used the SmartStart program to visit and work with Chris Eliasmith and Terrence C. Stewart at the Centre for Theoretical Neuroscience University of Waterloo (Canada). And I did my bachelors in Biology at the Free University of Berlin and the University of Vienna.

Email  /  Google Scholar  /  Github  /  LinkedIn  /  Twitter

The human brain is the most sophisticated and complex system in the universe.

Neural system identification aims at learning the response function of neurons to arbitrary stimuli using experimentally recorded data, but typically does not leverage normative principles such as efficient coding of natural environments. Visual systems, however, have evolved to efficiently process input from the natural environment. Here, we present a normative network regularization for system identification models by incorporating, as a regularizer, the efficient coding hypothesis, which states that neural response properties of sensory representations are strongly shaped by the need to preserve most of the stimulus information with limited resources. Using this approach, we explored if a system identification model can be improved by sharing its convolutional filters with those of an autoencoder which aims to efficiently encode natural stimuli. To this end, we built a hybrid model to predict the responses of retinal neurons to noise stimuli. This approach did not only yield a higher performance than the “stand-alone” system identification model, it also produced more biologically-plausible filters. We found these results to be consistent for retinal responses to different stimuli and across model architectures. Moreover, our normatively regularized model performed particularly well in predicting responses of direction-of-motion sensitive retinal neurons. In summary, our results support the hypothesis that efficiently encoding environmental inputs can improve system identification models of early visual processing.

Inter-experimental variability is commonly overlooked in systems neuroscience. We show how this leads to problems in the case of retinal cell type classification. As a solution we propose a flexible model, based on recent machine learning advances in domain adaptation, that removes this variability to unmask biological effects.

Elucidating causal, neurobiological underpinnings of behaviour is an ultimate goal of every neuroscientific study. However, due to the complexity of the brain as well as the complexity of the human environment, finding a~causal architecture that underlies behaviour remains a~formidable challenge. In this manuscript, we review the logical and conceptual issues with respect to causal research in neuroscience. First, we review the state of the art interventional and computational approaches to infer causal brain-behaviour relationships. We provide an~overview of potential issues, flaws, and confounds in these studies. We conclude that studies on the causal structure underlying behaviour should be performed by accumulating evidence coming from several lines of experimental and modelling studies. Lastly, we also propose computational models including artificial neuronal networks and simulated animats as a~potential breakthrough to causal brain-behaviour investigations.

* equal contribution. Thanks to Jon Barron for this amazing website template.