Divisive inhibition in recurrent networks.

Area of Science:

• Computational neuroscience
• Neural network modeling

Background:

• Models of the visual cortex propose that recurrent networks operating at high gain can generate response selectivity.
• However, high-gain recurrent networks present challenges, including operating near instability, sensitivity to synaptic changes, and slow responses to dynamic stimuli.

Purpose of the Study:

• To investigate a novel mechanism for enhancing the stability and dynamic response of recurrent neural networks in the visual cortex.
• To address the limitations of high-gain recurrent networks using divisive inhibition.

Main Methods:

• Simulated recurrent neural networks incorporating divisive inhibition.
• Analysis of network stability, response gain, and selectivity under varying stimulus conditions.
• Modeling the role of interneurons with divisive inhibition.

Main Results:

• Divisive inhibition effectively stabilizes high-gain recurrent networks, preventing runaway excitation.
• Networks with divisive inhibition exhibit faster responses to rapidly changing stimuli compared to standard high-gain networks.
• Network selectivity for stimuli is preserved or even enhanced by the proposed divisive inhibition mechanism.

Conclusions:

• Divisive inhibition, particularly when applied to interneurons within the network, provides a robust solution to the instability and sluggishness of high-gain recurrent models.
• This mechanism allows for the benefits of high gain, such as enhanced selectivity, while overcoming critical operational limitations in visual cortex models.