Interview with Prof. Dr. Gregor Volberg

I work at the Chair of General Psychology I and Methodology, Institute of Psychology.

In the bachelor’s course of Psychology I teach subject-specific methods. This includes experimentation techniques and statistical methods for evaluating experimental data. In the master’s course of Psychology I teach methods for processing and analyzing neuroscientific data. This includes a seminar on evaluating EEG data with Matlab and a seminar on modeling behavioral data with R.

One of my seminars for master’s students is about formal models of human cognition and behavior. The goal of modeling is to understand the processes that underlie human performance. This differs from typical machine learning applications. Example topics are manual voting reactions and categorizations.

The topics of my master’s seminars are new to most students. But we work with experimental data in both seminars, which is something that psychology students are very used to. The general procedure for adapting formal models of cognition (specify model – estimate parameters – determine model quality) is also similar to conventional models of parametric statistics, which the students are already familiar with.

In both courses with scripting languages, students learn basic programming techniques. In the seminar “Modeling behavioral data”, students get to know various formal models for describing cognition and behavior, e.g. diffusion models. They learn how to adjust model parameters and determine the model quality. The developed code can later easily be adapted to own questions. In the seminar “Analysis of EEG data with Matlab”, students learn the procedure for analyzing electroencephalographic data, from preprocessing to presentation of results.

In the neurosciences, especially in imaging, there is a trend towards increasingly abstract data descriptions. One no longer only looks at focal brain activity, but rather at entire activity patterns over time. Machine learning algorithms are suitable for identifying activity patterns of certain task or stimulus situations. The students need a basic knowledge of these methods in order to understand current research results.

I work on topics in visual neuroscience, primarily with EEG and behavioral data. I am interested in how separately represented visual information is summarized in the perception of an object (e.g. different features such as color and shape of the same stimulus, or stimulus features at different places). A particularly interesting phenomenon is synesthesia, in which, in addition to the normal perception, another perception experience occurs. For example, there are people who always perceive the letter “M” as being colored red. An international multicenter study on the role of language in synesthesia is currently under way, and I will examine people with German as their mother tongue.

There is a current example from my research on synesthesia. The assignment of colors to graphemes is individually different for people with synesthesia. For example, while one person sees the letter “M” in red, another person sees the letter in yellow. I was interested if similar letters have similar colors within one person, for example whether the letters “M” and “N” were both seen in red. For this I used distance matrices for letters consisting of artificial neural networks, which map the similarity of the representations in different structures of the visual cortex. I also created distance matrices for the colors, converted both distance matrices into a spatial representation and mapped the representations onto one another using special optimization processes. The correspondence between the two representations (shapes and colors) should of course be particularly high for those brain structures that trigger synaesthetic color perception. So far, the results show a heterogeneous picture with individually good fits between color and shape representations in different brain structures. With the large data set from the multicenter study, as described above, the individual differences can possibly be clarified.

During my studies in Bochum, I attended courses of the postgraduate programme in “Neuroinformatics”. Since then I have been interested in the simulation of human cognitions in technical systems. However, my knowledge of these topics has largely grown from the requirements of scientific work in self-study.

I use AI-related methods for evaluating experimental data and for special hypothesis testing. In this case of use, it goes very well. However, I find conventional statistics easier to use and use them more often.

I’m interested in modeling as a tool for empiricists. Formal models force you to be explicit and precise.

There are a number of informal collaborations at the own institute. In addition, there are collaborations, also informal, on tinnitus (MedBo) and on the analysis of multimodal neuroscientific data (machine learning group at the University of Regensburg). A official cooperation on synesthesia exists with a colleague from the University of Amsterdam.

The subject of machine learning will be prominently represented in the new Faculty of Computer Science at the University of Regensburg. The topic may be difficult to access for students from non-technical disciplines. I would like it if there was also a low-threshold offer of courses for these students.

I think the idea of ​​jointly presenting activities on the subject of AI/machine learning is nice. I am currently not involved in the projects listed on the website.

As an empiricist, I would like to point out that big data as a raw material for machine learning first has to be generated and made available. The willingness to do this can be promoted through open science initiatives in the specialist disciplines.


Mr. Volberg, thank you for taking the time to answer these questions. We wish you a nice day!