How AI Learns to Smell with Alex Wiltschko - #771

According to the TWIML AI Podcast, researchers are trying to give computers a sense of smell. Alex Wiltschko, founder of a company called Osmo, explains the challenge: mapping the messy relationship between a molecule's structure and how it actually smells, using the hundreds of receptors in the human nose as inspiration. His team uses graph neural networks and embedding spaces to group scents into perceptual neighborhoods and predict how a given molecule will smell. To train these models, Osmo built what it says is the largest proprietary olfactory dataset from scratch. The longer-term ambition reaches well past fragrance, toward things like disease detection, emotion sensing, and consumer devices that can, in effect, sniff.

Listen to this story

Hear this and more stories in a personalized audio briefing.

Open The Chonkerton