By Super Nervous
Summary: After thousands of years of accumulation, mankind has developed a variety of delicious flavors, but our taste buds will never be satisfied. An AI developer at Google uses AI to develop new dessert recipes in order to explore new possibilities. Will AI be more imaginative than humans in the field of recipe development?
Keywords: AI baking cooking recipe development
Baked foods have always occupies an important position among the various cuisines in the world. Soft bread, delicate cakes, and crisp biscuits are all irresistible.
Baking is not only a cooking technique, it is more like an art. Mix flour, eggs, butter, sugar and other raw materials in different proportions. After a series of physical and chemical reactions, snacks with different tastes and textures will be obtained, like exquisite works of art.
All kinds of desserts give people a sense of happiness in terms of taste and vision
Today, baking masters have developed dozens of classic desserts, such as egg tarts, puffs, and puff pastry, but in order to bring new experiences and excitement to the taste buds of foodies, they are still constantly researching and developing new products. .
As a baking enthusiast, Google AI engineer Sara has been exploring new possibilities. She combines this hobby with work and uses AI to develop new baking recipes.
Machine learning to do baking, develop two new products in minutes
Like many people, Sara, an AI engineer at Google Cloud, devoted a lot of time to the kitchen during the period when she was forced to stay home due to the epidemic.
Sara likes to combine work with hobbies and explore interesting things
Sara likes baking, but she found that when most people do baking, they search for some ready-made recipes on the Internet and then do them step by step. Although such an approach is safe, it has great limitations and it is difficult to innovate. But if you match them at random, you are likely to overturn the car, wasting materials and time.
Therefore, it is important to understand the scientific principles behind the baking recipe, so that you can understand the different effects of each ingredient and different proportions, and get rid of the limitation of a fixed recipe.
As an AI engineer, Sara believes that this task is very suitable for machine learning. "Train the machine learning model with existing data, let it master the laws, and then create the new formula we want."
Based on this idea, Sara quickly built an AI model. After the model learns 600 kinds of baking recipes, for the input recipes, it can accurately determine whether the baking result is bread, cake or cookie.
For a given formula, the AI model can accurately predict the result
Next, Sara let the AI model who has mastered the secret of the baking ingredient mix to create a cake and cookie mixture recipe, which she named "cakie" (cake+cookie).
The AI model lived up to expectations and generated an accurate formula according to Sara's needs. Sara conducted personal experiments and found that under the guidance of this new recipe, the baked "cakie" was very much in line with expectations, and the taste was very nice.
New desserts: bread cookies, both the fluffy feeling of cakes and the crispness of cookies
Later, Sara asked the AI model to create a recipe for "bread cookies", "breakie" (bread+cookie), and the results were also satisfactory to her.
A new variety of snacks that look like bread and biscuits
AI Baker: Well versed in the principle of dim sum matching
Sara introduced the construction process of this model in detail on the blog, let us see how AI can become a professional baker.
Data set collation
First, Sara and her colleagues collected more than 600 recipes from the Internet to form a recipe data set, including bread, cakes and biscuits. Then, the commonly used core raw materials are extracted, a total of 16 kinds, including flour, yeast, milk, water, salt, eggs, etc.
Then the unit of measurement of various ingredients in these formulas is unified. For example, some use "cup" as the unit, and some use "spoon" as the unit. The author converts all of them into "ounces" (1 ounce ≈ 28.35 grams). ).
Unify the units for the raw materials of each formula
Build a model and learn the formula
They used Google's AutoML Tables to build a classification model.
After creating a new tabular model, you can directly import data from csv, Google Sheets, or BigQuery databases. After the data is imported, you can see them in the "Training" tab:
After training the model with these data, the model has mastered the recipe characteristics of each bakery product, so that it can make more accurate predictions.
Analytical model interpretability
Through the analysis, Sara has further understood the basis of the AI model's judgment when making predictions. The results show that for the AI model, the importance of each ingredient in the baking recipe for decision-making is ranked as follows:
Important indicators on which the model predicts: butter, sugar, yeast and eggs
Of course, in fact, the recipes of various desserts are very complicated, and the above indicators are not fixed. For example, Sara analyzed the prediction results of a certain "cake" formula and found that eggs, butter, and baking soda were important indicators for AI to predict.
The model not only gives the judgment result, but also gives the basis for decision-making
In fact, in the field of baking, professionals have already compiled books on the principles of baking such as "Understanding the "Why" of Bread in Scientific Ways" and "Bread Bible", but for amateurs, there may not be enough time or Be patient to delve into it.
The AI bakers help us save this step. You don’t need to master scientific principles. You can also let AI help you create desserts that suit your taste. Isn’t it beautiful?
Is AI more reliable than humans in developing new dishes?
Tired of eating conventional dishes, more and more people have begun to pursue innovative dishes in recent years. However, the reality is that if you are not careful, innovative dishes will become daunting dark dishes: stir-fried oranges with greens, peppers with mooncakes, and bananas with watermelons...
Before AI learned to develop baking recipes, he had already set foot in the field of recipe research and development. Will it be more reliable than human chefs, innovating while avoiding rollovers?
In 2019, British pie maker Piglet's Pantry partnered with Esme Loans, a commercial lending platform, to allow the algorithm to learn thousands of existing British pies recipes (a text totaling nearly 1 million characters), and then learn to invent thousands A new pie recipe.
Later, after manual experimentation and improvement, five new pie recipes were selected for production. After some customers tried them, they said they were delicious.
AI-developed curry chicken pie
However, AI with too much imagination and too much innovation will inevitably make mistakes.
Earlier, even an AI that learned 30,000 recipes still failed to master the knack of combining various foods, and generated some recipes that seemed unbearable.
For example, the combination of blueberry + spinach + feta, and the combination of bacon + avocado + peach, etc...
A curious foodie baby said that the taste is hard to say
It seems that AI's skills in developing recipes are not stable enough, and sometimes even random combinations. Therefore, the majority of foodies still need to be mentally prepared to hand over the task of innovating dishes to AI, which saves time and also takes certain risks.
news source:
https://cloud.google.com/blog/topics/developers-practitioners/baking-recipes-made-ai
-- Finish--
Scan the QR code to join the discussion group
Get more high-quality data sets
Understand the application of artificial intelligence
Follow Top Meetings & Papers
Reply to "reader" for details
More exciting content (click on the picture to read)