All of human cooking compressed into 2 megabytes (arxiv.org)

Researchers have unveiled Epicure, a cutting-edge artificial intelligence project that distills the essence of human cooking into just two megabytes. By leveraging a multilingual corpus of 4.14 million recipes in seven languages, Epicure produces innovative ingredient embeddings using a skip-gram model retrained from scratch. The approach normalizes various ingredient strings into 1,790 canonical entries and creates complex ingredient-ingredient and ingredient-compound graphs, setting the stage for nuanced flavor interactions. With three Metapath2Vec variants that explore distinct relationship spectra—from chemical properties to recipe context—Epicure represents a novel intersection of culinary science and machine learning. This advancement is significant for the AI/ML community as it not only streamlines the understanding of culinary data through compact embeddings but also facilitates richer applications in areas like recipe generation, ingredient substitution, and flavor prediction. The structured graphs and statistical modeling introduce a level of precision that could revolutionize both consumer cooking experiences and professional culinary techniques. As Epicure bridges traditional cooking knowledge with AI, it opens new avenues for research, enhancing the way we conceptualize food pairings and recipes in the digital age.