US 11,862,300 B1
Formulation graph for machine learning of chemical products
Alix Schmidt, Midland, MI (US); and Ian Clark, Midland, MI (US)
Assigned to Dow Global Technologies LLC, Midland, MI (US)
Filed by Dow Global Technologies LLC, Midland, MI (US)
Filed on Feb. 27, 2023, as Appl. No. 18/114,398.
Int. Cl. G06N 3/08 (2023.01); G16C 20/70 (2019.01)
CPC G16C 20/70 (2019.02) 26 Claims
OG exemplary drawing
 
1. A method, comprising:
creating a digital formulation graph comprising a plurality of nodes and a plurality of edges based on a formulation for a chemical product, wherein:
a root node represents the chemical product;
a remainder of the plurality of nodes each represent a respective ingredient of the formulation; and
each of the plurality of edges exists between a respective parent node and a respective child node;
providing a respective embedding vector for each respective node;
inputting the digital formulation graph and embedding vectors to a graph neural network (GNN) trained to produce a feature vector for the digital formulation graph based on the embedding vectors and an architecture of the GNN;
inputting the feature vector to a supervised machine learning model trained to predict an attribute value of the chemical product based on the feature vector; and
receiving a prediction of the attribute value of the chemical product from the supervised machine learning model.