Discovering plasticity rules that organize and maintain neural circuits

Area of Science:

• Computational Neuroscience
• Systems Neuroscience
• Machine Learning

Background:

• Brain's intrinsic dynamics can enhance learning by providing a scaffold for task-aligned activity.
• Sequence generation is a key neural motif, exemplified by the zebra finch HVC song circuit.
• Maintaining these dynamics requires robustness against biological noise and perturbations.

Purpose of the Study:

• To discover a local plasticity rule that organizes and sustains sequence-generating neural dynamics.
• To ensure these dynamics are robust to perturbations like synaptic turnover and biological noise.
• To investigate the role of inhibitory plasticity in sequence generation.

Main Methods:

• A meta-learning approach was used to parameterize and optimize learning rules.
• Candidate rules were simulated in random networks and evaluated for their ability to encode time.
• Biological noise, including synaptic turnover, was introduced to test rule robustness.
• The impact of inhibitory plasticity was explored alongside excitatory plasticity.

Main Results:

• Meta-learning identified a temporally asymmetric rule, a generalization of Oja's rule, that organizes sparse sequential activity.
• The learned rule incorporated homeostasis, improving sequence maintenance under synaptic turnover.
• Plasticity rules adjusting both excitation and inhibition outperformed those acting solely on excitation.
• Learned plasticity effectively shaped excitatory cell dynamics for timing representations.

Conclusions:

• A novel, robust local plasticity rule was discovered that organizes and maintains sequence-generating neural dynamics.
• This rule demonstrates enhanced stability and recovery from perturbations compared to existing methods.
• Incorporating inhibitory plasticity is crucial for effectively shaping neural timing representations.