Smell the future: Machine olfaction is paving the way for digitizing odors

Machine learning has paved the way for voice and facial recognition but researchers are still struggling to quantify the elusive and sometimes vague sense of smell.

Machine olfaction, the automated simulation of the sense of smell, is an emerging field that uses robots or other automated systems to analyze air-borne molecules. Just like with sight and sound, machine learning is the key to digitizing smells because machines can learn to map molecular structures that create odors and translate them into a textual description, according to Ambuj Tewari, machine learning expert and Professor of Statistics at the University of Michigan.

“The machine learning model learns the words humans tend to use – for example, “sweet” and “dessert” – to describe what they experience when they encounter specific odor-causing compounds, such as vanillin,” Tewari writes for The Conversation.

Machine olfaction devices could also enable us to capture human odors and use them as biometric templates. In 2022, scientists from Japan’s Kyushu University developed a sensor called the “artificial nose” that can biometrically authenticate a person by the way their breath smells with an average accuracy of more than 97 percent.

Some companies are already banking on the new field. Machine olfaction startup Osmo received a US$3.5 million grant from the Bill & Melinda Gates Foundation last year to advance the company’s AI-enabled scent platform. Earlier in 2023, it also received US$60 million in Series A funding led by Lux Capital and Google Ventures.

The company, which was spun out of Google Research, wants to create a “map of odor” to predict what a molecule smells like from its structure. The platform will be used for creating compounds that repel, attract and even destroy disease-carrying insects such as mosquitos.

“Osmo’s science revealed a surprising link between insect and human olfaction, with our map of odor predicting how molecules smell to humans as well as insects,” says Osmo CEO Alex Wiltschko.

Quantifying smell, however, is a challenging task, and not just because odors can be difficult to describe. The internet has vast amounts of audio, image and video content that can be used by machine learning to train recognition systems. Machine olfaction has long faced a data shortage problem and without datasets, researchers had trouble training powerful machine learning models, according to Tewari.

A breakthrough was the 2015 DREAM Olfaction Prediction Challenge which invited research teams across the world to submit machine learning models. A research project called Pyrfume Project made more datasets publicly available.

A research team led by Osmo and the Monell Chemical Senses Center at the University City Science Center campus in Philadelphia was finally able to create an AI model that produced notable results in machine olfaction. This research has paved the way for odor prediction and digitizing odors.

The model can predict smell descriptions based on the structure of a molecule. It is based on a type of deep learning called graph neural networks and is trained on a dataset of 5,000 known odorants. The research study, which was published in Science in September 2023, found that AI outperformed individual human assessments for more than half (53 percent) of the molecules tested, according to Neuroscience News.

Article Topics

biometric template  |  biometrics research  |  body odor  |  machine learning  |  olfaction  |  Osmo