Dynamical mechanisms of how an RNN keeps a beat, uncovered with a low-dimensional reduced model

Area of Science:

• Neuroscience
• Computational Neuroscience
• Auditory Perception

Background:

• The neural basis for processing temporal patterns in music remains largely unknown.
• Understanding how the brain keeps time is crucial for explaining auditory perception.

Purpose of the Study:

• To investigate the neural mechanisms underlying temporal pattern perception in music.
• To model how the brain synchronizes to and anticipates rhythmic stimuli.

Main Methods:

• A biologically constrained recurrent neural network (RNN) with excitatory (E) and inhibitory (I) units was trained on a synchronization and continuation task across various tempos (2-8 Hz).
• A reduced three-variable rate model was developed to analyze the RNN's dynamic properties and oscillatory mechanisms.
• The model's dynamics were compared with neural recordings from monkeys performing the same task.

Main Results:

• The trained RNN generated a network oscillator whose frequency was controlled by an input current, replicating key neural dynamics observed in biological systems.
• Analysis of the reduced model confirmed the oscillatory mechanisms present in the full RNN.
• The neurally plausible model identified an E-I circuit with two distinct inhibitory sub-populations, one tightly synchronized with excitatory units.

Conclusions:

• The study elucidates a potential neural mechanism for temporal processing in rhythmic contexts, involving synchronized excitatory and inhibitory neural populations.
• The developed RNN and its reduced model provide a framework for understanding how the brain perceives and anticipates musical rhythms.
• These findings contribute to our understanding of the neural basis of auditory timing and music perception.