Max Planck Institute for Dynamics and Self-Organization -- Department for Nonlinear Dynamics and Network Dynamics Group
Personal tools
Log in

The pace of forgetting

here/Title
Date: 20.01.2011

Scientists from Göttingen are able to calculate for the first time how long neural networks in the cerebral cortex are capable of memorizing sensory information.

The dynamics of signal transfer in the brain are extremely chaotic, conclude scientists of the Max Planck Institute for Dynamics and Self-Organization (MPIDS), the University of Göttingen and the Bernstein Center for Computational Neuroscience Göttingen. Additionally, the scientists from Göttingen were able to calculate for the first time how quickly the information saved in the activity dynamics of neurons in the cerebral cortex decays. This "rate of forgetting" is surprisingly high - up to one bit per second and active neuron.

The brain encodes information as electric pulses - so called spikes - which are both sent and received by each of the roughly 100 billion interconnected neurons in the brain. A neuron collects every incoming electric pulse and under some circumstances passes on a pulse to its neighbors. This way every kind of information processed by the brain generates a specific activity pattern indicating which neuron passed on a spike to its neighbors at a specific point in time (fig. 1). This activity pattern can therefore be seen as a conversation log transcribing the communication between neurons.

How reliable is such a pattern? Do small changes in the neural conversation already generate a completely different pattern, just as if the change of a single word would change the whole meaning of a long discussion? Scientists call such behavior "chaotic". In this case, the dynamical processes in the brain would not be predictable over longer time periods. Furthermore, any information encoded in the activity patterns would quickly be lost due to small errors. On the other hand, a so called stable, non-chaotic dynamics would be a lot more resilient to errors. But the behavior of single neurons could no longer have any impact on the big picture.

The new results from Göttingen now show that the processes in the cerebral cortex, the main control center of the brain, are extremely chaotic. In order to come to this conclusion, it was crucial that the scientists for the first time use a realistic neuron model in their calculations. Whenever a spike reaches a neuron, additional electric potential is built up at its cell wall. It is only when this potential reaches a critical value that the neuron becomes active itself. "This event is of uttermost importance", says Prof. Dr. Fred Wolf from the MPIDS. "Without it, it is impossible to reproduce in calculations when exactly a neuron becomes active."

Previous models incorporated only highly simplified neuronal models and did not account for the precise requirements of spike generation. "This resulted in some cases of stable and some cases of chaotic dynamics", explains Michael Monteforte from the Max Planck Institute for Dynamics and Self-Organization and PhD student at the GGNB graduate school. The long-term disagreement whether the processes in the cerebral cortex are chaotic or not could not be explained this way.

With their sophisticated approach, the scientists from Göttingen could now calculate for the first time, how quickly activity patterns get lost due to minute changes. The decay rate is roughly one bit of information per second and neuron. "This extremely high erasement speed was very surprising for us" says Wolf. Apparently information gets lost in the brain roughly at the same speed as it can be "fed in" through sensory input.

This has fundamental consequences for our understanding of the neural code of the cerebral cortex. Due to the high decay rate, information on sensory input signals can only be maintained for a few spikes. The new results suggest that the dynamics of the cerebral cortex is tailored extremely well to the processing of short-time external snapshots.
 

Original publication: Michael Monteforte and Fred Wolf: Dynamical Entropy Production in Spiking Neuron Networks in the Balanced State, Physical Review Letters, 105, 268104, published Dec 31, 2010; DOI: 10.1103/PhysRevLett.105.268104
 

For additional information, please contact:

Dr. Birgit Krummheuer
Press Office
Max Planck Institute for Dynamics and Self-Organization
Tel.: +49 551 5176-668
Mobile: +49 173 3958625
 

Michael Monteforte
Max Planck Institute for Dynamics and Self-Organization, University of Göttingen and Bernstein Center for Computational Neuroscience Göttingen
Tel.: +49 551 5176-526
E-mail: follow this link

Prof. Dr. Fred Wolf
Max Planck Institute for Dynamics and Self-Organization, University of Göttingen and Bernstein Center for Computational Neuroscience Göttingen
Tel.: +49 551 5176-423
E-mail: follow this link