Google Researchers Training AI Models to Predict Smell

With the advancements in technology, AI has become good at seeing and listening like humans. But now tech giant Google is trying to develop another human sense for AI – smelling. The company is trying to develop a neural network that helps an AI identify the smell characteristics of a molecule.

According to Google, detecting smell is a multi-label classification problem which means a substance can have multiple smell characteristics. Lets take Vanillin, for example, a substance that is often used to develop an artificial vanilla flavor has multiple smell descriptors such as sweet, vanilla, and chocolate. Some characteristics are even stronger than others.

Therefore, in order to identify the smell profile of a molecule, researchers at Google used a graph neural networks (GNNs) which is a deep learning model that takes graphs as inputs. The team of researchers took the help of perfume experts to create labels of smell to commence the identification of a molecule's olfactory properties.

The neural network employed in this process starts the work by creating a representative vector using various properties such as atom identity and atom charge. It then broadcasts the vector to a neighboring node and then collectively passes to update function to get a vector for centered node.

The vector process is repeated for a layer and then continues for multiple layers till AI finally sums up or averages a vector for a molecule to identify multiple olfactory identifiers.

The researchers at Google asserted that the model outperforms older methods and it can be used to predict new or unclassified smells in RGB-layout like "odor embedding".

The team at Google used a set of around 5,000 molecules from perfumers who have expert noses. It later carefully matched each molecule with descriptions like "woody," "jasmine," or "sweet." Researchers used about two-thirds of the data set to train the network. They further tested whether it could predict the scents of the remaining molecules and it did.

In the coming future, the team of Google researchers wants to create solutions for digitalized scent creations. It also wants to build solutions for those without a sense of smell. Additionally, the wished to create more open datasets for research so researchers can leverage them for various scent-related machine learning models.

According to Alexei Koulakov, a researcher at Cold Spring Harbor Laboratory- "the project is valuable for introducing thousands of new molecules into the smell data sets, which are often relatively small, and that this data "could form the basis for improvements of this and other algorithms in the future."

He further said that "it's not clear if we can learn anything about human olfaction from a machine-learning model since the design of the neural network isn't the same as a human olfactory system."