Synthesizing network dynamics for short-term memory of impulsive inputs.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Systems Neuroscience

Background:

• The brain's ability to manage resources and coordinate neural networks is crucial for information processing.
• Understanding how neural dynamics encode, store, and recall information is a fundamental neuroscience challenge.
• Short-term memory relies on neural networks retaining representations of past stimuli.

Purpose of the Study:

• To investigate the dynamical mechanisms enabling neural networks to form and maintain short-term memory traces.
• To develop a model for how input gain adaptation contributes to memory persistence and information overwriting.
• To explore the relationship between memory policies and neural circuit function, particularly in the prefrontal cortex.

Main Methods:

• Utilizing computational modeling and control-theoretic approaches to analyze dynamical systems.
• Treating system state trajectories as abstract memory traces of prior inputs.
• Synthesizing variable input gain through optimization based on error minimization and overwriting prevention.

Main Results:

• A novel 'policy' for adapting input gain was derived, dependent on the network's current state.
• This policy establishes a recurrent neural network interaction between the policy and the underlying neural circuits.
• The findings align with theories suggesting prefrontal cortex subnetworks are dedicated to working memory.

Conclusions:

• The derived gain adaptation policy provides a mechanistic explanation for short-term memory persistence.
• The model highlights the interplay between dynamic gain control and information representation in neural networks.
• This work supports the hypothesis of specialized neural subnetworks for working memory within the prefrontal cortex.