US 11,699,076 B2
Method and system for learning rules from a data base
Csaba Domokos, Simmozheim (DE); Daria Stepanova, Renningen (DE); Jeremy Zieg Kolter, Pittsburgh, PA (US); and Po-Wei Wang, Pittsburgh, PA (US)
Assigned to ROBERT BOSCH GMBH, Stuttgart (DE); and CARNEGIE MELLON UNIVERSITY, Pittsburgh, PA (US)
Filed by Robert Bosch GmbH, Stuttgart (DE); and Carnegie Mellon University, Pittsburgh, PA (US)
Filed on Aug. 14, 2020, as Appl. No. 16/993,995.
Claims priority of application No. 19199308 (EP), filed on Sep. 24, 2019.
Prior Publication US 2021/0089894 A1, Mar. 25, 2021
Int. Cl. G06N 3/08 (2023.01); G06N 3/04 (2023.01); G06N 5/025 (2023.01)
CPC G06N 3/08 (2013.01) [G06N 3/04 (2013.01); G06N 5/025 (2013.01)] 10 Claims
OG exemplary drawing
 
1. A computer implemented method for learning rules from a data base, the data base including entities and binary relations between pairs of the entities, wherein each of the entities is either a non-numerical constant or a numerical value, and each binary relation of the binary relations between a non-numerical constant and a numerical value is a numerical binary relation and each binary relation of the binary relations between two non-numerical constants is a non-numerical binary relation, the method comprising the following steps:
deriving aggregate values from the numerical binary relations of the binary relations of the data base and/or the non-numerical binary relations of the binary relations of the data base;
deriving non-numerical binary relations from the aggregate values;
adding each of the derived non-numerical binary relations to the data base as a binary relation between two of the entities of the data base;
constructing differentiable operators, wherein each of the differentiable operators refers to a non-numerical binary relation of the data base or a derived non-numerical binary relation of the data base;
extracting rules from the differentiable operators using an artificial neural network based on neural logic programming (LP) framework, wherein the rules extracted from the differentiable operators are horn rules;
decoding the rules into decoded rules; and
predicting one or more relations between two or more entities of the data base by applying the rules and/or the decoded rules to the two or more entities of the data base.