On D-brief Leah Froats describes an enhancement to those algorithms which show you this based on that (known as the “nearest-neighbor search problem), which draws on as inspiration the neural architecture of fruit flies:
Fruit flies, however, have a mechanism in their brains that performs similarity searches in a very different way. Specifically, they expand the stimuli information, as opposed to compressing and simplifying it.
When fruit flies first sense an odor, 50 neurons fire in a combination unique to that smell. But instead of simplifying that information as computer programs would, the flies’ brains then send that information to a total of 2,000 neurons. With more neurons in play, the fly’s brain is able to give each smell a more unique label, meaning that it’s easier to categorize.
The flies then pare this information down to the top five percent or so of neural signals, effectively sorting out only the most salient information. This creates a pattern similar to a digital hash that the fly can then use to identify scents and respond accordingly.
When implemented on computers against a standard test set, it out-performed what I take to be current algorithms (her language is a bit confused). I’m having trouble understanding exactly how this all works, but it sounds fascinating. The Science article this is based on is behind a paywall, and I don’t currently have a subscription.
