Fruit fly-inspired AI learns smells quickly with far less memory

Researchers at OIST developed Spi-Fly, a neural network inspired by fruit fly olfactory systems, that learns odor categories from very few examples and can add new smells without forgetting old ones. It uses sparse coding and associative learning, avoiding backpropagation, which allows it to operate with low-precision weights and significantly reduced memory. Experiments show it excels in few-shot learning but lags after extensive training. The system has potential for real-world odor sensing applications but needs further testing on mixed odors and hardware integration.

(Source：The Brighter Side of News)