FlavorGraph helps incorporate food ingredients

After AI has entered games, autonomous driving, and other fields with mixed success, it is now entering the field of cooking, as Sony has developed a deep learning system called FlavorGraph designed to connect ingredients like garlic, olives and milk.

Overall, FlavorGraph predicts the pairing compatibility of two components by combining information about the molecules in a specific ingredient with the way people have used that ingredient in the past.

Coupling components

These suggestions can be used to predict relationships between compounds and foods, and the goal is to develop a smart deep learning model that recommends pairing complementary and new ingredients to help chefs with new creations.

Sony and Korean University researchers note that chefs have discovered how to incorporate ingredients through intuition, which led to the gradual development of ingredients such as cheese, tomatoes, apples, garlic and ginger.

The prevailing flavor

Many of these classic combinations were later explained by science, as researchers realized that ingredients that shared dominant flavor molecules often worked well together.

At the same time, other components that combine well may have completely different chemical formulations.

To find out why, the team examined both the molecular information about the ingredients and how they were historically used in recipes.


Hence they created a FlavorGraph database with flavor profiles, such as: bitter and sweet based on 1561 flavor molecules.

The team examined nearly a million recipes to see how the ingredients had been incorporated in the past.

The resulting data show the chemical compounds shared between foods and how they affect their overall taste, showing which foods can be matched with specific types of fruits.

There are some obvious examples of associated foods, such as cookies and ice cream, but others are less clear.


About the author

Leave a Reply

Your email address will not be published. Required fields are marked *