Dodo Pizza created pizza using neural networks and published the “source code” in the public domain

The system developed a pizza recipe based on the molecular compatibility of the ingredients.

“Open Source” Pizza Photo “Dodo Pizza”

Updated at 21:08: a comment by the chef of “Dodo Pizza” about the taste of “OpenSource” has been added.

Dodo Pizza developed OpenSource Pizza using neural networks. The system selected the ingredients based on molecular compatibility, and the “source code” of the pizza was published on GitHub. About this TJ told the company.

A pizza created by neural networks includes ten ingredients. Among them are both traditional ones like tomato sauce, chicken, cherry tomatoes, broccoli and mozzarella, and unusual ones, for example, melon, pear, tuna, mint and granola.

At Dodo, they worked on the system together with scientists from the Moscow Institute of Physics and Technology and Skoltech. The company noted that the neural networks analyzed more than 300 thousand recipes and a thousand ingredients, as well as the results of studies on the molecular compatibility of the ingredients.

The company noted that they created pizza as an experiment to check whether the neural network can create a delicious combination of ingredients. At the same time, developers faced several problems – for example, the lack of ready-made input data for analysis.

Generation of recipes for recipes is not the most common task, generation of pizzas is even more so. The most difficult was to collect the dataset – there are no such databases anywhere, you had to automatically process large culinary sites, and users write recipes on the sites, which causes all sorts of troubles like errors in the text.

Egor Baryshnikov

one of the developers of Dodo Pizza

In addition, neural networks required recipes in mathematical form. For this, the developers formed a set of ingredient vectors using the word2vec algorithm, which analyzes the occurrence of words in context.

New recipes were generated using two recurrent neural networks. First, the developers trained them on other recipes, and then added the molecular compatibility criterion – according to scientists, the ingredients with the most similar molecular structure are best combined.

The chef of “Dodo pizza” Konstantin Oganezov tried “Openers” pizza and said that he liked the taste.

The taste of pizza is hard to describe, but I would say that it looks like hot granola with caramelized fruits and tuna. It’s tasty.

Konstantin Oganezov

Chef Dodo Pizza

According to the developers, the resulting neural network can create both classic recipes and “crazy” ones. The source code for the project and the OpenSource pizza recipe were published on GitHub under a Creative Commons Attribution 4.0 license. This means that it can be freely used, including for commercial purposes.

Dodo Pizza will also prepare a limited batch of OpenSense pizza in the amount of 500 pieces. They can be tasted on October 10 in Moscow at a pizzeria on 46 Novoslobodskaya.

Cooking Open Source Pizza Photo Dodo Pizza

Back to top button