Sign In

Communications of the ACM

ACM TechNews

AI Rivals Human Nose When Naming Smells

View as: Print Mobile App Share:
To help test a new artificial intelligence nose, Jonathan Deutsch of Drexel University has spent hours sniffing and describing the odors of unknown chemicals.

The program, a so-called graph neural network, is excellent at imitating human sniffers, at least when it comes to simple odors.

Credit: DeAndra Forde

Researchers at artificial intelligence (AI) company Osmo, working with colleagues at Philadelphia’s Drexel University and the Monell Chemical Senses Center, developed a graph neural network that reliably matched human volunteers' identification of 55 odors, then predicted the smells of 500,000 additional molecules without having to produce or sniff them.

The researchers fed the structures and odor descriptions of 5,000 molecules to an AI to teach it to identify patterns in the training data by correlating a molecule's odor with attributes of its underlying atoms.

After calculating average human odor identification ratings, the researchers found the neural network got closer to this average than any individual in the volunteer group did in over half the cases.

The AI then deduced how the 500,000 hypothetical chemical structures should smell.

From Science
View Full Article - May Require Paid Subscription


Abstracts Copyright © 2023 SmithBucklin, Washington, D.C., USA


No entries found

Sign In for Full Access
» Forgot Password? » Create an ACM Web Account