High accuracy single-layer free-space diffractive neuromorphic classifiers for spatially incoherent light.

Free-space all-optical diffractive systems have shown promise for neuromorphic classification of objects without converting light to the electronic domain. While the factors that govern these systems have been studied for coherent light, the fundamental properties for incoherent light have not been addressed, despite the importance for many applications. Here we use a co-design approach to show that optimized systems for spatially incoherent light can achieve performance on par with the best linear electronic classifiers even with a single layer containing few diffractive features. This performance is limited by the inherent linear nature of incoherent optical detection. We circumvent this limit by using a differential detection scheme that achieves greater than 94% classification accuracy on the MNIST dataset and greater than 85% classification accuracy for Fashion-MNIST, using a single layer metamaterial.

Single dendrite-targeting interneurons generate branch-specific inhibition.

Microcircuits composed of dendrite-targeting inhibitory interneurons and pyramidal cells (PCs) are fundamental elements of cortical networks, however, the impact of individual interneurons on pyramidal dendrites is unclear. Here, we combine paired recordings and calcium imaging to determine the spatial domain over which single dendrite-targeting interneurons influence PCs in olfactory cortex. We show that a major action of individual interneurons is to inhibit dendrites in a branch-specific fashion.

A general principle of neural arbor branch density.

The tree-like structures of a neuron that are responsible for distributing (axons) or collecting (dendrites) information over a region of the brain are called arbors. The size of the territory occupied by an arbor and the density of the arbor branches within that territory are important for computation because these factors determine what fraction of a neural map is sampled by a single cell and at what resolution [1]. Arbor territory size and branch density can vary by many orders of magnitude; however, we have identified a universal relationship between these two physical properties revealing a general neural architectural design principle. All of the arbors (axons and dendrites) we have studied (including fish retinal ganglion cells, rodent Purkinje cells, and the cortical arbors of various neural classes from rat, cat, monkey, and human) are found to be systematically less dense when they cover larger territories. This relationship can be described as a power law. Of several simple biological explanations explored, we find that this relationship is most consistent with a design principle that conserves the average number of connections between pairs of arbors of different sizes.