The Convallis rule for unsupervised learning in cortical networks.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Machine Learning

Background:

• Cortical synaptic plasticity's mechanisms are known, but its computational role in sensory representation is unclear.
• Understanding how plasticity shapes neural representations is crucial for brain function and artificial intelligence.

Purpose of the Study:

• To introduce and validate the Convallis rule, a novel framework for cortical synaptic plasticity.
• To investigate the computational principles underlying sensory representation development in the neocortex.

Main Methods:

• Developed the Convallis rule based on unsupervised learning and constrained optimization.
• Implemented the rule in a simulated spiking neural network processing speech stimuli.
• Compared the Convallis rule's performance with spike-timing-dependent plasticity (STDP) using in vitro experimental data.

Main Results:

• The Convallis rule enabled a recurrent network to develop rate representations of speech, facilitating classification.
• The rule reproduced and extended experimental results beyond standard STDP.
• STDP alone demonstrated inferior learning performance compared to the Convallis rule.

Conclusions:

• The Convallis rule offers a potential computational principle for neocortical plasticity.
• The rule's mathematical form aligns with experimental evidence of dual coincidence detection mechanisms.
• This framework bridges normative, phenomenological, and mechanistic insights into cortical function.