Luke’s paper “Temporal prediction captures retinal spiking responses across animal species” is now out in Nature Communications!

https://www.nature.com/articles/s41467-026-77399-y

The retina’s function has long been framed as one of two extremes: an efficient compressor of visual input, like a camera, or a predictor of future stimuli. We built a spiking neural network model of the retina and trained it on natural movies under metabolic-like constraints, to either encode the present or predict the future.

When optimized to efficiently predict ~100 ms ahead, the model not only reproduces retina-like receptive fields and their mosaic organization, but also captures a range of hallmark retinal phenomena: latency coding, motion anticipation, differential tuning, and stimulus-omission responses.

Strikingly, the temporal-prediction model also accurately predicts how retinal ganglion cells respond to natural images and movies across multiple animal species. These findings suggest the retina isn’t just compressing visual input, but rather that it is fundamentally organized to give the brain foresight into the visual world.