Retinal Spiking Neural Networks Learn the Ropes of Spacetime and Master the Vision

Abstract

Purpose : Eyes can actively and clearly catch each photon event in the changing visual field, which make the brain be on the lighted scene accurately and instantly. This task requires the retina is high organized to pipeline the multiple dimensions (x, y, z, t) overview of events. The mathematics and physics of spacetime events should explain the virtual vision in retinal neural networks.

Methods : The parallel and hierarchical pathways of retina were proved in patch clamp experiments. Spacetime coordinate transformation is used to analyze the histological structures and electrical signals of retinal visual system. The zero-spacetime-interval projection of photons mediates the virtual reality of retinal neural network.

Results : Retina locates the events rotating the single eye, two eyes, head and body at the different frequency series. The translation, rotation, composition transforms of spacetime coordinates are involved in the procedure. There are three types of cones (red, green, blue) and three kinds of photoreceptors: cones, rods and ipRGCs to form their parallel and hierarchical networks where they establish the virtual frequency series of the real events in the visual field. The whole-cell patch clamp experiments displayed the retinal structure and computation in the cellular level of retinal neurons, their discrete/integral frequency series cover all the real events. The parallel and hierarchical networks of bipolar and ganglion cells reproduce the virtual reality with event-evoked Ca²⁺/Na⁺ spikes. Retinal neural network is operating with Ca²⁺ spikes in bipolar cells and Ca²⁺/Na⁺ spikes in ganglion cells. The sustained Ca²⁺ currents is the rotation of time dimension such as the rod system. Time dimension decides the transmission of double paths in the spatial pathways, which produces the ON/OFF pathways in the virtual spacetime. The ON/OFF and Ca²⁺/Na⁺ spike signals form the transfer and impulse functions with the event timing. All pathways are superposed to reestablish the multiple-dimension reality, covering the continuous spacetime of visual field.

Conclusions : The superposing parallel pathways of retina build the virtual reality of visual field. The spacetime explanation of retinal neural networks could guide the artificial intelligent (AI) vision in the brain-inspired direction.

This abstract was presented at the 2024 ARVO Annual Meeting, held in Seattle, WA, May 5-9, 2024.