Formula and recipe generation with feedback loop

Abstract

Techniques to mimic a target food item using artificial intelligence are disclosed. A formula generator is trained using ingredients and using recipes and, given a target food item, determines a formula that matches the given target food item. A flavor generator is trained using recipes and their associated flavor information and, given a formula, the flavor generator determines a flavor profile for the given formula. The flavor profile may be used to assist the formula generator in generating a subsequent formula. A recipe generator is trained using recipes and, given a formula, determines a cooking process for the given formula. A food item may be cooked according to a recipe, and feedback, including a flavor profile, may be provided for the cooked food item. The recipe and its feedback may be added to a training set for the flavor generator.

What is claimed is: 
     
         1 . A method comprising:
 generating, using an artificial intelligence formula model and an artificial intelligence flavor model, a predicted formula including ingredients that together mimics a target food item,   wherein the artificial intelligence formula model is trained to obtain a latent space with a probability distribution of first sets of ingredients of a plurality of first recipes,   wherein the artificial intelligence flavor model is trained to obtain a function mapping second sets of ingredients of a plurality of second recipes to flavor profiles;   generating, using an artificial intelligence recipe model, a predicted recipe for the predicted formula,   wherein the artificial intelligence recipe model is trained to learn relationships between at least ingredients and cooking steps of the plurality of first recipes;   receiving feedback data for a cooked food item associated with the predicted recipe, wherein the feedback data includes a flavor profile of the cooked food item,   wherein the artificial intelligence flavor model is retrained using the predicted recipe and the feedback data.   
     
     
         2 . The method of  claim 1 , wherein ingredients of the predicted formula are generated by decoding a latent code sampled from the latent space, wherein the latent space is sampled according to user-provided control definitions, wherein the user-provided control definitions include one or more of ingredient restriction definitions, flavor profile definitions, or nutritional feature definitions. 
     
     
         3 . The method of  claim 1 , wherein the predicted formula mimics the target food item with regards to nutritional features and flavor features. 
     
     
         4 . The method of  claim 1 , wherein the artificial intelligence flavor model comprises a classifier architecture, wherein the artificial intelligence recipe model comprises an autoregressive language model architecture, wherein an output of the artificial intelligence flavor model guides the artificial intelligence recipe model in generating a subsequent candidate formula. 
     
     
         5 . The method of  claim 1 , wherein generating the predicted recipe comprising sequentially generating a plurality of actions, wherein a first action of the plurality of actions is based on the predicted formula, and each subsequent action of the plurality of actions is based on one or more previous actions. 
     
     
         6 . The method of  claim 5 , wherein at least one the one or more previous actions is user-modified. 
     
     
         7 . The method of  claim 1 , wherein each set of the first sets of ingredients is associated with a recipe from a plurality of first recipes, and wherein each set of the second sets of ingredients and a respective flavor profile are associated with a recipe from the plurality of second recipes. 
     
     
         8 . The method of  claim 1 , wherein the artificial intelligence formula model comprises a first autoencoder and a second autoencoder, wherein the latent space is obtained by the second autoencoder, wherein the latent code is decoded based on a particular encoded representation, of the first sets of ingredients, generated by the first autoencoder model. 
     
     
         9 . The method of  claim 1 , wherein artificial intelligence flavor model comprises:
 a certainty level classifier that generates, for each flavor category of a plurality of flavor categories, a certainty level to indicate a level of certainty that a flavor associated with that flavor category is present in a particular set of ingredients;   a plurality of flavor predictors associated with the plurality of flavor categories, wherein each flavor predictor generates a deeper level of flavor granularity corresponding to an associated flavor category.   
     
     
         10 . The method of  claim 1 , wherein the predicted formula comprises at least a set of one or more ingredients that mimics the target food item with regards to nutritional features. 
     
     
         11 . One or more non-transitory computer-readable storage media storing one or more instructions programmed for generating a candidate set of ingredients, when executed by one or more computing devices, cause:
 generating, using an artificial intelligence formula model and an artificial intelligence flavor model, a predicted formula including ingredients that together mimics a target food item,   wherein the artificial intelligence formula model is trained to obtain a latent space with a probability distribution of first sets of ingredients of a plurality of first recipes,   wherein the artificial intelligence flavor model is trained to obtain a function mapping second sets of ingredients to flavor profiles;   generating, using an artificial intelligence recipe model, a predicted recipe for the predicted formula,   wherein the artificial intelligence recipe model is trained to learn relationships between at least ingredients and cooking steps of a plurality of first recipes;   receiving feedback data for a cooked food item associated with the predicted recipe, wherein the feedback data includes a flavor profile of the cooked food item;   wherein the artificial intelligence flavor model is retrained using the predicted recipe and the feedback data.   
     
     
         12 . The one or more non-transitory computer-readable storage media of  claim 11 , wherein the latent space is sampled according to user-provided control definitions, wherein the user-provided control definitions include one or more of ingredient restriction definitions, flavor profile definitions, or nutritional feature definitions. 
     
     
         13 . The one or more non-transitory computer-readable storage media of  claim 11 , wherein the predicted formula mimics the target food item with regards to nutritional features and flavor features. 
     
     
         14 . The one or more non-transitory computer-readable storage media of  claim 11 , wherein generating the predicted recipe comprising sequentially generating a plurality of actions, wherein a first action of the plurality of actions is based on the predicted formula, and each subsequent action of the plurality of actions is based on one or more previous actions. 
     
     
         15 . The one or more non-transitory computer-readable storage media of  claim 11 , wherein the artificial intelligence flavor model comprises a classifier architecture, wherein the artificial intelligence recipe model comprises an autoregressive language model architecture, wherein an output of the artificial intelligence flavor model guides the artificial intelligence recipe model in generating a subsequent candidate formula. 
     
     
         16 . 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:
 generating, using an artificial intelligence formula model an artificial intelligence flavor model, a predicted formula including ingredients that together mimics a target food item, 
 wherein the artificial intelligence formula model is trained to obtain a latent space with a probability distribution of first sets of ingredients of a plurality of first recipes, 
 wherein the artificial intelligence flavor model is trained to obtain a function mapping second sets of ingredients to flavor profiles; 
 generating, using an artificial intelligence recipe model, a predicted recipe for the predicted formula, 
 wherein the artificial intelligence recipe model is trained to learn relationships between at least ingredients and cooking steps of a plurality of first recipes; 
 receiving feedback data for a cooked food item associated with the predicted recipe, wherein the feedback data includes a flavor profile of the cooked food item; 
 wherein the artificial intelligence flavor model is retrained using the predicted recipe and the feedback data. 
   
     
     
         17 . The computing system of  claim 16 , wherein the latent space is sampled according to user-provided control definitions, wherein the user-provided control definitions include one or more of ingredient restriction definitions, flavor profile definitions, or nutritional feature definitions. 
     
     
         18 . The computing system of  claim 16 , wherein the predicted formula mimics the target food item with regards to nutritional features and flavor features. 
     
     
         19 . The computing system of  claim 16 , wherein generating the predicted recipe comprising sequentially generating a plurality of actions, wherein a first action of the plurality of actions is based on the predicted formula, and each subsequent action of the plurality of actions is based on one or more previous actions. 
     
     
         20 . The computing system of  claim 16 , wherein the artificial intelligence flavor model comprises a classifier architecture, wherein the artificial intelligence recipe model comprises an autoregressive language model architecture, wherein an output of the artificial intelligence flavor model guides the artificial intelligence recipe model in generating a subsequent candidate formula.