How Seashells Can Teach Us About Neural Networks

Image: Alistair Boettiger

The meticulous design of a seashell sure is pretty to you and me, but when UC Berkeley biophysicist George Oster looked at one, he saw something else entirely: neural networks. Here’s why:

During shell construction, the mantle is always extended just a bit beyond the lip of the shell, inspecting its prior handiwork; Oster and Ermentrout hypothesized that pigment patterns from days past are scanned and interpreted by the mantle’s nerve network, triggering waves of excitation and inhibition that yield detailed instructions for the next round of construction. “What the mantle is doing is ‘tasting’ back in time,” says Oster, “so it can predict what it should do the next day and so that the pattern will be continuous.”

By charting these discrete patterns of neural excitation and inhibition, Oster and Ermentrout were able to build a mathematical model for shell formation that accounts for virtually any design observed in nature, from the zigzagging lines of Natica communis to the seemingly random patterns of mottled patches on a cone snail’s shell. “A single equation is sufficient to explain this tremendous diversity of patterns,” says Alistair Boettiger, a Berkeley graduate student who developed a computational modeling program for Oster and Ermentrout based on their findings. The team has modeled more than 30 shell types, and in each case the simulation bears a striking resemblance to the real thing. The program is even able to compensate for changes in growth and patterning caused by scratches and scrapes picked up in a mollusk’s tumultuous life at sea.

Michael Eisenstein of Seed Magazine has more: Link