Hebbian Crosstalk Prevents Nonlinear Unsupervised Learning

TL;DR

This paper investigates how synaptic crosstalk affects nonlinear Hebbian learning in neural networks, revealing a critical threshold beyond which learning performance drastically deteriorates, with implications for cortical learning processes.

Contribution

It demonstrates that crosstalk causes a sudden performance drop in nonlinear Hebbian learning models, highlighting a critical threshold relevant to cortical learning mechanisms.

Findings

Performance drops sharply at a critical crosstalk level

Crosstalk impacts nonlinear ICA learning

Implications for cortical learning processes

Abstract

Learning is thought to occur by localized, experience-induced changes in the strength of synaptic connections between neurons. Recent work has shown that activity-dependent changes at one connection can affect changes at others (crosstalk). We studied the role of such crosstalk in nonlinear Hebbian learning using a neural network implementation of Independent Components Analysis (ICA). We find that there is a sudden qualitative change in the performance of the network at a critical crosstalk level and discuss the implications of this for nonlinear learning from higher-order correlations in the neocortex.