Fruit-fly-inspired spiking network Spi-Fly learns smells with few samples and little memory

Brighter-Side-News · reddit · 2026-09-22

A fruit fly brain has only about 140,000 neurons yet recognizes odors remarkably efficiently. Researchers at the Okinawa Institute of Science and Technology (OIST) borrowed that strategy to build Spi-Fly, an algorithm based on a spiking neural network modeled on the fruit fly olfactory system.

Designed for settings with limited training examples, memory and compute, it performed especially well when only a few examples of each odor were available, and learned new categories without substantially forgetting earlier ones.

Original post →

More from Research

Research channel →