Favoring Generalists over Specialists: How Attentional Biasing Improves Perceptual Category Learning

A model of cortical lea.rning is proposed, \vhieh incorpora.tes supervised feedback using two forms of attention: (i) feature-specific attention which allmvs the nct.vwrk to learn associations between specific feature conjunctions (or categories) and out.puts, and (ii) nonspecific att.entionaJ "vigilance'' which biases this learning when the associations appear to be incorrect. Attentional vigilance. improves lea.ming if it f<lvors, via. lcamcd modulatory weights, qeneralist categories over specialist categories. A biologically plausible neural network i:::; proposed which implements these cornputational principles and which outperforms r-;evcra.l a.It.crnative classifiers on classification bcnchmarkr-;.