A multi-compartment spiking neural network model demonstrates that biologically feasible deep learning can be achieved if sensory inputs and higher-order feedback are received by different dendritic compartments.

The medial prefrontal cortex and orbitofrontal cortex work in concert to match state representations from feedback to those at choice, and the strength of these common neural codes predict credit assignment precision.

What can artificial intelligence learn from neuroscience, and vice versa?.

A landmark-based cross-modality alignment method robust to variation in landmark sets is applied to annotate an EM time series of Caenorhabditis elegans embryonic development as a community resource.

Models that generate tandem alignments of cell polarities are more readily compatible with the formation of PIN1 polarity patterns in plant leaf buds than the most widely accepted “up-the-gradient” model.

Novel evidence for a role of feedback in the perception of uniform surfaces in the human brain suggests that feedback already re-enters at an early visual processing stage.

The assignment of borders to foreground objects occurs in cortical columns in primate visual cortex, and first in deep layers, suggesting a central role for feedback.

The synaptic structure in mouse V1 is explained by a synergy of homeostatic plasticity in incoming and outgoing synapses of inhibitory interneurons, establishing a stimulus-specific balance of excitation and inhibition.

A biologically plausible learning rule enables recurrent neural networks to model the way in which neural circuits use supervised learning to perform time-dependent computations.

Running guinea fowl maintain stable running after loss of the stretch reflex in a major ankle extensor muscle, by increasing feedforward muscle activation to maintain ankle stiffness and work output.