Fruit Fly-Inspired Odor Classification Algorithm Shows Top Performance With Few Training Samples

Researchers at the Okinawa Institute of Science and Technology Graduate University developed “SpyFly,” an odor classification algorithm modeled on how fruit flies process smells. The algorithm models the fruit fly olfactory system as a three-layer neural network and applies a sparse-coding method that activates only a small number of neurons. When tested on two odor databases, the researchers found that it achieved top performance in few-shot learning, accurately classifying odors with a small number of training samples.