Implementing the Brain's Learning Mechanism via the Axon Framework

We present a new framework that satisfies the three criteria a learning model of the neocortex should meet: computational suitability, algorithmic implementability, and neurochemical implementation detail.

Currently, the only framework that satisfies all these criteria is error-driven predictive learning driven by the corticothalamic circuit, which is based on a competitive kinase synaptic plasticity mechanism.

This mechanism was implemented through the Axon neural simulation framework, which uses spiking neurons, and its learning capability was demonstrated across various cognitive tasks. It is expected to serve as a future alternative that could replace back propagation, dramatically reducing learning time.