Useful dynamic regimes emerge in recurrent networks
Srdjan Ostojic, Ecole Normale Superior, Paris :: Nature Neuroscience, Vol. 17, No. 4, April 2014 :: www.nature.com/neuro
Summary and review of the above paper
Time-varying patterns of spatiotemporal activity in the brain underly most human activity. Increases in synaptic strength are seen as producing a shift from a small computational potential to high computational potential. Excitation and inhibition cancels out most of the time; however, irregular spikes persistently arise due to voltage fluctuation, and because of this cortical neurons during stationary periods have neurons firing irregularly. As synaptic strengths for both excitation and inhibition increase, so firing rates vary more in time, and this produces the transition from a homogeneous to a heterogeneous regime.
The initial homogeneous stage of firing is indicated to be suitable for transmitting information about external inputs. Over any period of time, such a system will simply increase or decrease its firing rate in direct relation to the external input. Such a system is relaying information without adding anything to it.
The distinct heterogeneous regime is a function of underlying instability in firing, with small fluctuations leading to differences in neuronal firing rates that generate further fluctuations in their turn. Such systems do not return to a stable state. Instead the firing rates of individual neurons in such a system fluctuate markedly between different times and between different neurons.
This less regular or heterogeneous pattern is argued to provide a basis for more complex information processing and learning. Transmission of information is somewhat degraded in this system, but against this, there is a capacity for the non-linear processing of stimuli. Moreover, the repeated presentation of the same stimulus does not necessarily produce the same response because of the sensitivity of this system to both initial conditions and subsequent noise.