All of human cooking compressed into 2 megabytes

Researchers have created Epicure, a machine learning model that compresses culinary knowledge from 4.14 million multilingual recipes into ingredient embeddings, normalizing ingredients into 1,790 canonical entries across seven languages. The model uses three variants of neural networks that balance recipe co-occurrence patterns with chemical flavor compounds to map relationships between food ingredients. This work demonstrates how AI can efficiently represent the complexity of global cooking traditions in a compact format suitable for computational analysis.