Diversity in prediction error signals across neocortical laminae - Blake Richards

Blake Richards

McGill University

"Diversity in prediction error signals across neocortical laminae"

Join Zoom

Abstract
Computational neuroscientists have postulated that the neocortex implements hierarchical predictive coding, whereby each region sends predictions of the inputs it expects to lower-order regions, allowing the latter to efficiently propagate prediction errors only. This hypothesis is supported by the presence of prediction-error-like responses to unexpected visual flow stimuli in the somatic compartments of layer 2/3 (L2/3) pyramidal neurons in primary visual cortex (VisP). However, to date, very few studies have investigated other types of prediction error in this circuit, or whether other potential sites for prediction-error signals exhibit similar responses, such as L5 pyramidal neurons, or pyramidal neuron apical dendrites in L1. Here, we addressed this using 2-photon calcium imaging in the VisP of awake mice, in a passive viewing task. We investigated the responses of L2/3 and L5 pyramidal neuron somata and apical dendrites to two stimuli: 1) a visual flow stimulus punctuated by unexpected disruptions, wherein a subset of the stimulus temporarily reversed direction, and 2) sequences of random Gabor patches where unexpected orientation distributions were occasionally inserted. We found that both L2/3 somata and apical dendrites exhibited prediction-error-like signals to the unexpected visual flow events, independent of visual flow direction, whereas L5 somata and apical dendrites did not. In contrast, both L2/3 and L5 dendrites showed responses to unexpected Gabor images, but only after multiple sessions of exposure. Our results suggest that there is no single "prediction error" response in VisP, and different compartments compute different forms of prediction error. This may indicate multiple local cost functions in the neocortical circuit.