Neural capacitance: neural network selection via edge dynamics

A non-transitory computer readable medium comprising computer executable instructions which when executed by a computer cause the computer to perform the method of:

• removing an output layer from a pre-trained neural network model;
• incorporating a neural capacitance probe (NCP) unit with multiple layers on top of one or more bottom layers of the pre-trained neural network model;
• randomly initializing the neural capacitance probe (NCP) unit;
• training a modified neural network model by fine-tuning the one or more bottom layers on a target dataset for a maximum number of epochs, the modified neural network model comprising the neural capacitance probe (NCP) unit incorporated with multiple layers on top of the one or more bottom layers of the pre-trained neural network model;
• obtaining an adjacency matrix from the initialized neural capacitance probe (NCP) unit;
• computing a neural capacitance metric using the adjacency matrix;
• selecting an active model using the neural capacitance metric;
• and
• configuring a machine learning system using the active model.

An apparatus comprising:

• a memory;
• and
• at least one processor, coupled to said memory, and operative to perform operations comprising:
• removing an output layer from a pre-trained neural network model;
• incorporating a neural capacitance probe (NCP) unit with multiple layers on top of one or more bottom layers of the pre-trained neural network model;
• randomly initializing the neural capacitance probe (NCP) unit;
• training a modified neural network model by fine-tuning the one or more bottom layers on a target dataset for a maximum number of epochs, the modified neural network model comprising the neural capacitance probe (NCP) unit incorporated with multiple layers on top of the one or more bottom layers of the pre-trained neural network model;
• obtaining an adjacency matrix from the initialized neural capacitance probe (NCP) unit;
• computing a neural capacitance metric using the adjacency matrix;
• selecting an active model using the neural capacitance metric;
• and
• configuring a machine learning system using the active model.