>>11187018>A computer calculating at gorillions of times a second reached the same optimization that billions of years of biological trial-and-error didthat's not exactly what this does. it's a model for how your eyeballs hook up when you're a fetus.
it's a neat paper, but i'd like to see more evidence that it's an effective feature extractor and if/how different dynamical systems affect its final quality.
if the "wave generator" only creates simple, low-level patterns then i'm not convinced it could learn a complex hierarchy of low-to-high level features.
i bet there will be at least a couple papers in the next year or so that study how different inputs affect the final topology and feature quality. real videos, artificial "noise-driven spatiotemporal wave generators", or some mix. wish i could do it, but i'm tired of dicking around with neural networks.
