Systems and methods to mimic target food items using artificial intelligence
Systems and methods to mimic a target food item using artificial intelligence are disclosed. The system can learn from open source and proprietary databases. A prediction model can be trained using features of the source ingredients to match those of the given target food item. A formula comprising a combination of most relevant source ingredients and their proportions can be determined using the trained prediction model. A set of existing recipes can be used as a dataset to train a recurrent neural network (RNN) and/or other suitable models. The RNN can be used to determine a recipe to mimic the target food item. The recipe may comprise a cooking process for the set of ingredients in the formula and can be cooked by a chef. The recipe may be further modified as necessary based on human feedback on sensorial descriptors.
1. A computer-implemented method to generate a recipe using plant-based ingredients to mimic a given animal-based food item, the method comprising:
receiving matching plant-based source ingredients from a plurality of plant-based source ingredients that match an animal-based target food item in a feature space of the plurality of plant-based source ingredients;
applying a machine learning generator model to the at least the matching plant-based source ingredients to generate a recipe that uses the matching plant-based source ingredients to mimic the animal-based target food item, the recipe comprising a cooking process,
wherein the machine learning generator model is trained using a training set, the training set comprising a set of processed existing recipes stored in an existing recipes database, each existing recipe in the set of processed existing recipes including a list of ingredients and a list of instructions for cooking the ingredients in the list of ingredients.
2. The computer-implemented method of claim 1 , further comprising:
retrieving, from a target database, a set of data features associated with the animal-based target food item, the set of data features for the animal-based target food item comprising at least one of:
physiochemical data features of the animal-based target food item,
nutritional data features of the animal-based target food item, or
molecular data features of the animal-based target food item;
retrieving, from a source ingredients database, a respective set of data features for each of the plurality of plant-based source ingredients, the respective set of data features for each of the plant-based source ingredients comprising at least one of:
physiochemical data features of the plant-based source ingredient,
nutritional data features of the plant-based source ingredient, or
molecular data features of the plant-based source ingredient;
creating a different training set from the respective set of data features for each of the plurality of plant-based source ingredients;
training a machine learning prediction model by using the different training set to generate the feature space of the plurality of plant-based source ingredients.
3. The computer-implemented method of claim 2 , further comprising:
executing a computer-implemented feature compression method on the respective set of data features for each of the plurality of plant-based source ingredients, to determine a compact representation of the respective set of data features for each of the plurality of plant-based source ingredients, wherein the different training set comprises the compact representation of the respective set of data features for each of the plurality of plant-based source ingredients.
4. The computer-implemented method of claim 2 , wherein the set of data features associated with the animal-based target food item and the respective set of data features for each of the plurality of plant-based source ingredients are each associated with a same set of data feature types, and wherein the same set of data feature types comprises at least one of:
a nutritional descriptor feature type,
a physiochemical descriptor feature type, or
a molecular descriptor feature type.
5. The computer-implemented method of claim 2 , wherein the machine learning prediction model is based on gradient boosting trees, and wherein the machine learning generation model is based on a recurrent neural network.
6. The computer-implemented method of claim 1 , wherein the cooking process includes a list of actions to be performed using the matching plant-based source ingredients.
7. The computer-implemented method of claim 1 , further comprising:
representing each source ingredient of the plurality of plant-based source ingredients in a D-dimensional space based on the respective set of data features identified for the source ingredient, the representation of the source ingredient in the D-dimensional space comprising at least one of: physiochemical, nutritional, or molecular descriptors of the source ingredient.
8. The computer-implemented method of claim 1 , further comprising:
determining a formula and a corresponding similarity score, wherein the formula combines the-matching plant-based source ingredients in specific proportions and the similarity score indicates how similar the formula is to the animal-based target food item in the feature space, wherein determining the formula includes executing an optimization process to determine the specific proportions for combining the matching plant-based source ingredients in the formula using least absolute shrinkage and selection operator (Lasso) regression.
9. The computer-implemented method of claim 1 , wherein the training set is created by:
obtaining a set of existing recipes, each existing recipe comprising a respective list of ingredients, respective quantities and respective instructions for cooking the respective list of ingredients; and
obtaining the set of processed existing recipes based on modifying each of the existing recipes, in the set of existing recipes, to include respective atomic directions for cooking the respective list of ingredients of the existing recipe.
10. One or more non-transitory computer-readable storage media storing one or more instructions programmed for generating a recipe using plant-based ingredients to mimic a given animal-based food item, when executed by one or more computing devices, cause:
receiving matching plant-based source ingredients from a plurality of plant-based source ingredients that match an animal-based target food item in a feature space of the plurality of plant-based source ingredients;
applying a machine learning generator model to the at least the matching plant-based source ingredients to generate a recipe that uses the matching plant-based source ingredients to mimic the animal-based target food item, the recipe comprising a cooking process,
wherein the machine learning generator model is trained using a training set, the training set comprising a set of processed existing recipes stored in an existing recipes database, each existing recipe in the set of processed existing recipes including a list of ingredients and a list of instructions for cooking the ingredients in the list of ingredients.
11. The one or more non-transitory computer-readable storage media of claim 10 , wherein the one or more instructions, when executed by the one or more computing devices, further cause:
retrieving, from a target database, a set of data features associated with the animal-based target food item, the set of data features for the animal-based target food item comprising at least one of:
physiochemical data features of the animal-based target food item,
nutritional data features of the animal-based target food item, or
molecular data features of the animal-based target food item;
retrieving, from a source ingredients database, a respective set of data features for each of the plurality of plant-based source ingredients, the respective set of data features for each of the plant-based source ingredients comprising at least one of:
physiochemical data features of the plant-based source ingredient,
nutritional data features of the plant-based source ingredient, or
molecular data features of the plant-based source ingredient;
creating a different training set from the respective set of data features for each of the plurality of plant-based source ingredients;
training a machine learning prediction model by using the different training set to generate the feature space of the plurality of plant-based source ingredients.
12. The one or more non-transitory computer-readable storage media of claim 11 , wherein the one or more instructions, when executed by the one or more computing devices, further cause:
executing a computer-implemented feature compression method on the respective set of data features for each of the plurality of plant-based source ingredients, to determine a compact representation of the respective set of data features for each of the plurality of plant-based source ingredients, wherein the different training set comprises the compact representation of the respective set of data features for each of the plurality of plant-based source ingredients.
13. The one or more non-transitory computer-readable storage media of claim 11 , wherein the set of data features associated with the animal-based target food item and the respective set of data features for each of the plurality of plant-based source ingredients are each associated with a same set of data feature types, and wherein the same set of data feature types comprises at least one of:
a nutritional descriptor feature type,
a physiochemical descriptor feature type, or
a molecular descriptor feature type.
14. The one or more non-transitory computer-readable storage media of claim 11 , wherein the machine learning prediction model is based on gradient boosting trees, and wherein the machine learning generation model is based on a recurrent neural network.
15. The one or more non-transitory computer-readable storage media of claim 10 , wherein the cooking process includes a list of actions to be performed using the matching plant-based source ingredients.
16. The one or more non-transitory computer-readable storage media of claim 10 , wherein the one or more instructions, when executed by the one or more computing devices, further cause:
representing each source ingredient of the plurality of plant-based source ingredients in a D-dimensional space based on the respective set of data features identified for the source ingredient, the representation of the source ingredient in the D-dimensional space comprising at least one of: physiochemical, nutritional, or molecular descriptors of the source ingredient.
17. The one or more non-transitory computer-readable storage media of claim 10 , wherein the one or more instructions, when executed by the one or more computing devices, further cause:
determining a formula and a corresponding similarity score, wherein the formula combines the-matching plant-based source ingredients in specific proportions and the similarity score indicates how similar the formula is to the animal-based target food item in the feature space, wherein determining the formula includes executing an optimization process to determine the specific proportions for combining the matching plant-based source ingredients in the formula using least absolute shrinkage and selection operator (Lasso) regression.
18. The one or more non-transitory computer-readable storage media of claim 10 , wherein the training set is created by:
obtaining a set of existing recipes, each existing recipe comprising a respective list of ingredients, respective quantities and respective instructions for cooking the respective list of ingredients; and
obtaining the set of processed existing recipes based on modifying each of the existing recipes, in the set of existing recipes, to include respective atomic directions for cooking the respective list of ingredients of the existing recipe.
19. A computing system comprising:
one or more computer systems comprising one or more hardware processors and storage media; and
instructions stored in the storage media and which, when executed by the computing system, cause the computing system to perform:
receiving matching plant-based source ingredients from a plurality of plant-based source ingredients that match an animal-based target food item in a feature space of the plurality of plant-based source ingredients, wherein the feature space is generated by a machine learning prediction model;
applying a machine learning generator model to the at least the matching plant-based source ingredients to generate a recipe that uses the matching plant-based source ingredients to mimic the animal-based target food item, the recipe comprising a cooking process,
wherein the machine learning generator model is trained using a training set, the training set comprising a set of processed existing recipes stored in an existing recipes database, each existing recipe in the set of processed existing recipes including a list of ingredients and a list of instructions for cooking the ingredients in the list of ingredients.
20. The computing system of claim 19 , wherein each of the plurality of plant-based source ingredients is represented in a D-dimensional space based on their respective set of features comprising at least one of physiochemical descriptors, nutritional descriptors, or molecular descriptors.