Leveraging deep learning to control neural oscillators.

Area of Science:

• Computational Neuroscience
• Artificial Intelligence in Medicine
• Neural Engineering

Background:

• Modulating neural oscillator firing times is crucial for treating neurological conditions like Parkinson's disease, Tourette's syndrome, and epilepsy.
• Neural desynchronization, shifting oscillators from in-phase to out-of-phase firing, is a key control objective.
• Existing methods for optimizing neural stimuli often require system simplification or complete parameter knowledge, limiting real-world application.

Purpose of the Study:

• To develop a trained artificial neural network capable of accurately predicting the effects of square-wave stimuli on neurons.
• To utilize the neural network for solving complex neural control problems, specifically focusing on desynchronization.
• To enable effective neural desynchronization with minimal output information from the neuron.

Main Methods:

• Training an artificial neural network to estimate the impact of square-wave stimuli on neural oscillators.
• Employing a machine learning approach that requires minimal output data from the neural system.
• Applying the trained network to control tasks involving neural desynchronization.

Main Results:

• The artificial neural network accurately estimates the effects of stimuli on neurons.
• The network successfully facilitates the desynchronization of pairs of neurons.
• The approach enables the creation of clustered neuronal subpopulations even with coupling and noise present.

Conclusions:

• A novel artificial neural network approach significantly advances the ability to control neural oscillator timing.
• This method overcomes limitations of traditional techniques by requiring less system information.
• The findings offer a promising tool for developing new therapeutic strategies for neurological disorders and enhancing learning paradigms.