Learning shapes cortical dynamics to enhance integration of relevant sensory input

Adaptive sensory behavior is thought to be dependent on processing in recurrent cortical circuits, but how dynamics in these circuits shape the integration and transmission of sensory information is not well understood. Here we study neural coding in recurrently connected networks of neurons driven by sensory input. Here they show analytically how information available in the network output varies with alignment between feedforward input and the integrating modes of the circuit dynamics. 

Taking into consideration this theory:

• It is  analyzed neural population activity in the visual cortex of mice that learned to discriminate visual features
• It is also found that over learning, slow patterns of network dynamics realigned to better integrate input relevant to the discrimination task.
• And this realignment of network dynamics could be even explained by the changes in excitatory-inhibitory connectivity amongst neurons turned to relevant features. 
• These results also suggest that learning tunes the temporal dynamics of cortical circuits to optimally integrate sensory input

Conclusion:

Sensory discrimination relies on the temporal integration of optimally weighted sensory input:

 It is firstly asked that the dynamic properties of recurrent networks influence their capacity to discriminate sensory inputs. This scenario was considered and also had one of two possible stimuli appear for the duration of its trial. To determine how these inputs should be integrated for optimal discrimination performance, they adopted a signal-processing perspective 

Two noisy stimuli can optimally discriminate from instantaneous sensory input to the network by taking a one-dimensional linear combination of the inputs to various types of neurons[as shown in the below image]weighted according to the "linear discriminate"....This is the linear combination of inputs that used to achieve the best comprises between separating the mean inputs under the two stimuli and avoiding projected noise [as shown in below figure  1D]

[Stimulus describes the national performance which is based on the temporal integration of weighted sensory input]

Results demonstrate that the network can generate distinct activity patterns in response to two different and also continuous stimuli if it temporarily integrates the input stimuli weighted according to their projection on optimal linear discriminant.

A: Feedforward inputs to two various neurons

B: A weighted sum

C: The temporarily integrated inputs projection for each stimulus 

D: Distributions of instatenuos feedforward for each of the two stimuli

E: The signal 

F: The instantaneous and temporarily integrated signal-to-noise ratio of these two projections