Jump to content
Powered by

Jens Kremkow – does the neural network work like a gate?

Billions of nerve cells in our brain communicate with each other and lead to phenomena such as thought, feeling and memory. What fundamental information-processing principles govern the “chaos” of electrical impulses? Dr. Jens Kremkow from the Bernstein Center in Freiburg creates computer models of neural processes. Kremkow and his team carried out many complex simulations from which they were able to develop simple models of the human brain and test how they control the transmission of electrical excitation rather like neuronal gates. In a recent set of simulations, Kremkow and his colleagues used layered cortex models, which led to the discovery of a factor that may well play a role in decision making. This factor is time.

Dr. Jens Kremkow © private

A pneumatic drill used to dig up streets, a TV programme, snippets of a conversation in a kitchen - these are just a few examples of the hundreds of stimuli we are exposed to at any given moment in our natural environment. Why is it that we only "hear" some of these stimuli whilst others completely pass us by. Selective perception is an example of decision-making processes on the neuronal level: different nerve cell networks in the human brain compete with each other; all this relates to which electrical activity is transmitted and which is inhibited, with the result that the inhibited one dies off. Researchers call the process of transferring or inhibiting activity "gating" because the human brain appears to function rather like a sluice gate in that it only allows certain information through. However, nerve cells and nerve cell assemblies do not recognise the content of stimuli; they do not "know" that something is red, loud or beautiful. So how does this process work on the neural network level? "In principle it all depends on the complex interactions of excitatory and inhibitory connections," said Dr. Jens Kremkow from the Bernstein Center Freiburg. "We will have to focus on the rules that govern the interaction of excitation and inhibition, which forms the basis for transmitting information".

Control from one layer to another

Kremkow has been working in the field of theoretical neurobiology since 2003. He was born in Hanover, Germany in 1979, studied biology in Freiburg and Montreal and subsequently went on to specialise in computer science and neurobiology. His degree thesis in 2005 involved the computer simulation of neural networks in order to find out how the layers of the human cerebrum are connected with each other. Kremkow participated in a binational doctoral programme in Marseille and Freiburg, receiving his doctorate in 2009. As part of the EU project “FACETS – Fast Analogue Computing with Emergent Transient States”, Kremkow dealt with the principles behind the processing of information in the visual system, focusing in particular on how the system processes visual stimuli and regulates the “gating” of information. Prior to this project, the majority of theories were developed from experiments that investigated artificial stimuli in experimental animals, using grids of black and white lines amongst other things. For example, cats with electrodes attached to their brain for recording information were shown such stimuli. “Our project partners also use natural stimuli, for example films of a walk through a park,” said Kremkow. “I attempted to programme models of neural networks and explain the data obtained with experimental animals in order to be able to describe the processes in human brains.”

Schematic of a layered model showing how the individual groups are connected with each other by excitatory and inhibitory inputs. © private

Kremkow and his colleagues constructed a network of neurons (layered networks) on the computer. An important aspect of the model is the property of nerve cells to influence the activity of other nerve cells, either in an excitatory or inhibitory manner. The researchers' simulations involved groups of nerve cells connected with each other by way of excitatory and inhibitory contacts (synapses). They based their developments on anatomical data: the excitatory synapses in the human brain form direct connections between two groups of neurons which then pass on two different signals. On the other hand, the inhibitory input is switched over by way of an interstation, in other words, the inhibitory interneurons. Therefore, in the layered model, the excitatory signals reach one level higher (downstream neurons) slightly before the inhibitory signals. In their simulations, the scientists found that this delay was of crucial importance for the gate neurons' "decision" to transmit the signals. "In principle, the transmission of information can be controlled in two ways," said Kremkow. One way is via the amplitude of the signals arriving from the upstream neurons, i.e. the strength of the input. If excitatory and inhibitory impulses arrive simultaneously, the question arises as to which of the two has a higher overall electrical activity. If the activity of excitatory neurons is higher than the activity of inhibitory neurons, the downstream neurons are excited. Once this excitation exceeds a certain threshold, the signal is then transferred to the next level, i.e. the "gate" neurons" which control the signals that are transmitted onward. If the inhibitory input is higher than the excitatory input, the activity dies off.

Red traffic lights ask you to stop; green ones allow you to cross the road

”Other scientists had previously shown that amplitude-dependent gating actually plays a role in the transmission of signals,” said Kremkow. “But this only works with slowly changing stimuli.” If the stimuli are very short-lived, for example like the firing of a starting gun for a 100 m race, it would appear that another principle plays a role in deciding which signal to transfer.

Kremkow was able to show that this depends on the length of the delay between the arrival of the inhibitory signals and the excitatory signals. If the delay is relatively long (measured in milliseconds), the excitatory signals have enough time to accumulate and cross the threshold at which point the neurons start to fire. If the delay is short, the inhibitory signals suppress the excitatory signals, which means that the activity of the cells in the “gate” is drowned too quickly for the signal to be propagated.

This finding supports Kremkow’s hypothesis that temporal gating can form the basis for the selection of one of several alternatives in the human brain. Kremkow initially described the principle on the theoretical level before sending a version of the model to his project partners in Paris who then tested it on cats. As expected, the model predicted a behaviour that corresponded well with the data obtained using experimental animals, both with artificial stimuli such as grids of black and white lines for which Kremkow had originally developed the model, as well as natural stimuli such as movies.

The traffic light in the human brain: Here the connections between different groups of neurons correspond to motorways and the overlap of excitatory and inhibitory signals is represented by the lights which control the transmission of excitatory signals. © private

In 2009, Kremkow returned to Freiburg to do his postdoctoral research. Kremkow and his colleagues developed models that combined the two principles of gating (amplitude and temporal delay) thus showing that this enables the brain to control the entire range of stimuli, including slow and rapid stimuli. The researchers went a step further still: they wanted to know whether their model was able to choose between possible actions at the single nerve cell level. To do this, they fed their layered cortex model with two different stimuli with the idea that nerve cells transmitted one stimulus whilst inhibiting the other stimulus. This is a relatively simple decision-making process. Can it be controlled by manipulating the time difference between the incoming excitatory and inhibitory signals? "The results clearly showed that our network worked basically like a set of traffic lights," said Kremkow. "If the time delay of the excitatory and inhibitory signals was set to zero, the activity of the cells was drowned. This can be compared to red traffic lights. If the delay is very short, there is a 50 per cent probability that a signal will be transmitted. This is like an amber traffic light, where some drivers stop and some do not. Only lengthy delays correspond to green traffic lights, enabling the signal to be transmitted without interruption."

Back to basics

It is still not known whether the principle of temporal delay also plays a key role in real decisions that are made by a real brain. There is some evidence from the field of attentiveness research that different factors, including differing environmental stimuli, can alter the delay time. Thus it can quite safely be assumed that “time” is at least one relevant factor. Kremkow and his colleagues from the University of Freiburg are now going a step further by testing the results in brain slices. Kremkow, his newborn son and his wife are set to move to New York where the biologist will initially go back one step in order to look for further fundamental principles of information processes. From March, Kremkow will be communicating with his Freiburg colleagues via ‘Skype’ or other online communication media.

Further information:

Dr. Jens Kremkow
Bernstein Center Freiburg and
Institute of Biology III
Neurobiology and Biophysics
University of Freiburg
Schänzlestr. 1
79104 Freiburg
Tel.: +49 (0)761/203-2861
Fax: +49 (0)761/203-2860
E-mail: kremkow(at)biologie.uni-freiburg.de

Website address: https://www.gesundheitsindustrie-bw.de/en/article/news/jens-kremkow-does-the-neural-network-work-like-a-gate