Browse all publications by topic
Browse all publications by year
- T. Hoch, G. Wenning, and
K. Obermayer. Calculating the Rate of Information Transmission through
Populations of Neurons.
.
In Soc. Neurosci. Abstr. 30, 2004.
CD-ROM.
Neurons in the cerebral cortex receive 5000 to 60000 synaptic
contacts from other neurons. Given that cortical neurons fire spontaneously
with a rate of 5 to 10 Hz, these neurons are exposed to a considerable amount
of synaptic activity. This background activity can significantly improve the
transmission of weak (sub-threshold) signals in neurons (Wenning and
Obermayer, PRL, 2003) which becomes optimal for a certain stimulus dependent
level of noise. Using a population of leaky integrate-and-fire neurons we
explore how information transmission depends on this background activity and
on the statistics of the input signals. Each neuron receives a common
band-limited aperiodic Gaussian stimulus embedded in white noise which is
independent across the neurons in the population. Information transmission is
then quantified by three different information-theoretic cost functions:
discriminability, the mutual information, and a reconstruction method with
linear filter (Bialek et al., Science, 1995). Numerical simulations show that
the optimal noise level does not depend on the number of neurons when
evaluated using the discriminability and the mutual information as a measure
of transmission. If information transmission, however, is quantified using
the reconstruction method, then the optimal noise level depends on the number
of neurons in the population. Shifting the frequency range of the stimuli to
higher values immediately reduces this dependency on the population size. We
will provide evidence that this behavior is due to the linear filter used for
stimulus reconstruction, which is not able to account well for the negative
part of a signal with low frequency range. Because natural stimuli often have
a strong low frequency component, the widely used reconstruction method with
linear filter may significantly underestimate the information rate in the
context of population coding.
|
