US 12,346,802 B2
Neuro-symbolic next state prediction based on multiple potential causes
Akifumi Wachi, Tokyo (JP); Ryosuke Kohita, Tokyo (JP); and Daiki Kimura, Tokyo (JP)
Assigned to International Business Machines Corporation, Armonk, NY (US)
Filed by International Business Machines Corporation, Armonk, NY (US)
Filed on Dec. 3, 2020, as Appl. No. 17/110,400.
Prior Publication US 2022/0180166 A1, Jun. 9, 2022
Int. Cl. G06N 3/08 (2023.01); G06N 3/04 (2023.01)
CPC G06N 3/08 (2013.01) [G06N 3/04 (2013.01)] 9 Claims
OG exemplary drawing
 
1. A computer-implemented method (CIM) performed by a hybrid computing system comprising a neural network that includes machine learning functionality and a symbolic rule based software module that includes a symbolic reasoner, wherein the CIM comprises steps of:
receiving a current state information set including a plurality of attribute values characterizing a current and historical status of an environment with the plurality of attribute values including at least a temperature attribute value characterizing a temperature value in the environment;
receiving first action information including information indicative of a first action being taken in the environment;
applying the neural network of the hybrid computing system to the current state information data set and the first action information that includes the information indicative of the first action being taken in the environment to make a first prediction of a future temperature value in the environment, with the neural network including a pre-trained multi-label classifier;
applying the symbolic reasoner of the hybrid computing system to the current state information data set to make a second prediction of the future temperature value in the environment, with the symbolic reasoner including a plurality of software based rules, with the application of the symbolic reasoner including applying each rule of the plurality of software based rules to a symbolic representation of the current state information; and
making a third prediction of the future temperature value based on the first prediction and the second prediction.