Mathematics of vision - Interview with Fred Wolf
You are the first German scientist that has been awarded the Mathematical Neuroscience Prize. How do you see the development of Computational Neuroscience in your country?
For the neurosciences as well as for other areas of the life sciences, mathematics and theory are becoming increasingly important. In Germany, the Bernstein Network has succeeded in creating a structure that makes it easy for young people to see this and to pursue interdisciplinary research interests.
One important factor is the network’s visibility. If you are interested in the brain, you will come across the Bernstein Network sooner or later. In addition, we have established master and PhD programs in which students can simultaneously follow their interests in biology, mathematics and computer sciences.
When I first became interested in the brain, I was perhaps the first physics student also taking neuroscience at the University of Frankfurt. At that time in Frankfurt, the physics curriculum was very open making this possible. Today, with the bachelor and master system, we have to find other ways to enable students to bridge the gap between mathematics and biology.
How did you personally benefit from the Bernstein Network?
None of the discoveries in my group would have been possible without the network. In particular, I value that it offers the possibility to contribute to problems beyond my main field of interest and to learn new things through collaborations – and I think many others feel the same. The Bernstein Network is a great community and researchers are very aware that together we can achieve more. I have had very successful collaborations with colleagues who have a very different approach to neuroscience. Without the network anybody would have to think twice whether or not to embark on such often risky projects.
One focus of your research is the question of how information is stored in neural circuits. What is the most far-reaching insight from this research area?
We have long assumed that the exact biophysical properties of individual neurons do not play a major role when it comes to the function of large networks. But when we took a closer look, we could not sustain that view.
A neuron in the cortex generates a spike maybe once per second on average. When we modified the spike generation mechanism just a little bit, the collective properties of the whole network could change radically even if the number of spikes is exactly preserved. It is now emerging that in different networks, neuronal properties are specifically designed to improve the information bandwidth in the network by up to a factor of ten. We would have never thought that one can put so much information into in neural spikes. The next step will be to investigate in more detail the biophysical properties behind this optimized performance - one can perhaps indeed say that a new research topic has emerged.
Another major topic of your research is the development of the visual cortex. Why the visual cortex in particular?
I am fascinated by the neocortex. The visual cortex is the part of the neocortex can be most easily approached with mathematical models, because it processes visual information and the visual world is dominated by Euclidean geometry. The neocortex has only evolved about 200 million years ago in mammals and it provides them with an extraordinary learning ability and flexibility. I am interested in how this brain structure can function like a computing machine that processes information while, at the same time, it is capable of learning and thus has to be able to restructure itself.
One question we have examined concerns the division of labor between the neurons of the visual cortex. How many neighboring neurons does each cell in the visual cortex "consult with" before it finally decides which aspects of the visual world it will process and how? We have investigated models in which we only allow contacts with e.g. the 100 or 1000 closest cells. These models, however, cannot explain reality. Only very large and wide-ranging networks provided correct predictions about the development of the cortex. These models are now viewed as the best evidence that the cerebral cortex is a large-scale self-organized system.
What are your plans for the future, which questions would you like to address?
I am particularly interested in addressing how a system as complex as the neocortex arose and changed in evolution. The architecture of the visual cortex - the way in which cells are arranged and connected that specialize in different aspects of visual processing - has emerged more than once in evolution, but always in the exact same form. Viewed from an evolutionary perspective, this raises many questions: What exactly is the advantage this architecture brings about? Can a network of this size and complexity be encoded in the genes and be inherited? And if so, how does biology nonetheless maintain its learning ability?
In addition, I am interested in how the cerebral cortex has been scaled up in evolution. The neocortex has grown enormously e.g. in the lineage leading to humans. To better understand how such a large-scale remodeling of neuronal circuits can happen in evolution would be really nice. Maybe theory-driven approaches can contribute here.
The interview was conducted by Katrin Weigmann.
Read the news item Fred Wolf receives the Mathematical Neuroscience Prize 2017