Inhibitory-stabilization is sufficient for history-dependent computation in a randomly connected attractor network

Area of Science:

• Computational neuroscience
• Neural circuit dynamics
• Cognitive function

Background:

• Effective information processing relies on stimulus and prior history.
• Existing attractor network models lack biologically realistic firing rates.
• Recurrent excitatory networks struggle with stable low firing rates.

Purpose of the Study:

• To demonstrate how inhibition-stabilized networks can achieve computational abilities of recurrent networks.
• To show stabilization at biologically realistic low firing rates.
• To explore the functional role of inhibitory stabilization in history-dependent computation.

Main Methods:

• Modeling randomly connected inhibition-stabilized attractor networks.
• Analyzing network activity and firing rates.
• Investigating the impact of inhibitory feedback on computation.

Main Results:

• Inhibition-stabilized networks preserve computational abilities of recurrent excitatory networks.
• Network activity stabilizes at arbitrarily low firing rates.
• Transient oscillations due to inhibitory feedback enable state-dependent responses.

Conclusions:

• Excitatory-inhibitory balance stabilizes neural network activity.
• Inhibitory stabilization is functionally important for history-dependent computation.
• Inhibition-stabilized dynamics may underlie cognitive tasks and cortical computation.