Normative Networks for Source Separation via Local Plasticity and Dendritic Computation

Researchers introduced Predictive Entropy Maximization, a blind source separation method built from a regularized entropy objective.

The model uses local feedforward and lateral plasticity rules, plus output constraints and dendritic-style error updates.

It achieved competitive results against stronger baselines, even on correlated sources and noisy mixtures.

The approach offers an online, biologically plausible alternative for separating mixed signals with limited global supervision.