Training dynamically balanced excitatory-inhibitory networks.

Area of Science:

• Computational neuroscience
• Neural network modeling

Background:

• Understanding neural circuit computation requires biologically plausible models.
• Building functional networks with distinct excitatory and inhibitory neurons obeying Dale's law is challenging.

Purpose of the Study:

• To develop a method for constructing balanced recurrent neural networks.
• To ensure models adhere to biological constraints like Dale's law.

Main Methods:

• Utilized a target-based approach combined with fast online constrained optimization.
• Developed models for both rate and spiking recurrent neural networks.

Main Results:

• Successfully built functional neural network models with balanced excitation and inhibition.
• Demonstrated that these balanced networks can be trained for complex temporal pattern generation and input-output tasks.
• Preserved biologically desirable features including Dale's law and response variability.

Conclusions:

• The target-based optimization approach is effective for creating biologically constrained neural network models.
• Balanced excitatory-inhibitory networks are capable of complex computations and retain biological realism.