Mars, a confectionery company, approached Google to create an AI-powered sweet dish based on Maltesers candy. Sara Robinson, a Google Cloud AI employee who loves baking at her leisure, created a neural network that was trained on recipes for sweet pastries, using over several hundred dishes. The neural network, after training, came up with a new cooking method that uses an unusual ingredient.
Google explained that since Maltesers are widespread in the British Isles, the new dessert was supposed to contain ingredients typical of British sweet baked goods. Considering this condition, Sarah Robinson, while writing and training the neural network, applied a dataset based on the methods of cooking British dishes. The dataset included 4 common baked goods in the UK: cakes, biscuits, rolls and pastries. Robinson used Keras’ TensorFlow API and used AI Platform Hyperparameter Tuning to optimize the AI model. After finding the perfect combination of hyperparameters, Sarah unrolled the AI model using AI Platform Prediction. The model was able to work out a list of the constituent ingredients, their volume and formed a hybrid dessert. This is how the Maltesers AI Cakes were created.