Back to the present: self-supervised learning in neocortical microcircuits

Sensory systems in the mammalian brain exhibit rich representations that ultimately support complex behaviours. Microcircuits across neocortical layers are believed to underlie the development of these representations, but exactly how this occurs remains unclear. In the canonical view of the neocortical microcircuit, first-order thalamic projections target layer 4 (L4), which in turn projects onto layer 2/3 (L2/3) pyramidal cells (PCs) and then finally onto layer 5 (L5) PCs. Although this is a motif found throughout the neocortex, the functional role of this architecture is not known. Here, inspired by recent observations showing that first-order thalamic input also targets L5 PCs, we introduce a model in which the canonical microcircuit enables the brain to learn through temporal self-supervision. In our model, L2/3 PCs learn to predict thalamic inputs by comparing past sensory information originating from L4 with the current thalamic input received by L5 PCs. First, we trained our model to predict upcoming visual input in a sequential task. When combined with top-down contextual input our model can successfully learn to predict visual input through local L2/3-L5 self-supervised learning. In addition, we show that this form of predictive learning leads to a network with sensory denoising properties. Next, we trained the model in visuomotor tasks in which the visual input is either linearly or non-linearly coupled to the animal's speed. After training, when halting the visual input, the network generates error signals that are reminiscent of mismatch responses observed experimentally in both superficial and deep layers. Overall, our work proposes that classical canonical microcircuit motifs underlies a form of self-supervised learning in the brain.