Teaching Signal Synchronization in Deep Neural Networks with Prospective Neurons

Area of Science:

• Neuroscience
• Computational Neuroscience
• Cognitive Science

Background:

• Working memory relies on neural persistence, but hierarchical organization causes timing delays.
• These delays disrupt learning by desynchronizing neural activity from teaching signals.

Purpose of the Study:

• To investigate how neurons with adaptive currents can compensate for timing delays in hierarchical networks.
• To demonstrate a mechanism for synchronizing neural activity with teaching signals for efficient learning.

Main Methods:

• Mathematical analysis of prospective coding in neurons with adaptive currents.
• Simulations of learning algorithms in hierarchical networks.
• Behavioral experiments on motor control tasks.

Main Results:

• Neurons with adaptive currents exhibit prospective coding, anticipating future inputs.
• This prospective coding synchronizes neural activity with teaching signals across various learning algorithms.
• Prospective coding enables learning and memory retrieval in networks with slowly integrating neurons.

Conclusions:

• Neural adaptation via adaptive currents offers a solution to critical timing problems in neural computation.
• Prospective coding facilitates efficient learning and memory formation in dynamic environments.