US 12,450,519 B2
Root cause analysis for deterministic machine learning model
James Charles Rohrkemper, Harbor Springs, MI (US); Richard Paul Sonderegger, Dorchester, MA (US); Anna Chystiakova, Sunnyvale, CA (US); Kenneth Paul Baclawski, Waltham, MA (US); Dieter Gawlick, Palo Alto, CA (US); Kenny C. Gross, Escondido, CA (US); Zhen Hua Liu, San Mateo, CA (US); and Guang Chao Wang, San Diego, CA (US)
Assigned to Oracle International Corporation, Redwood Shores, CA (US)
Filed by Oracle International Corporation, Redwood Shores, CA (US)
Filed on Aug. 31, 2021, as Appl. No. 17/462,592.
Prior Publication US 2023/0061280 A1, Mar. 2, 2023
Int. Cl. G06N 20/00 (2019.01); G06F 16/242 (2019.01)
CPC G06N 20/00 (2019.01) [G06F 16/242 (2019.01)] 20 Claims
OG exemplary drawing
 
1. A non-transitory computer-readable medium comprising instructions which, when executed by one or more hardware processors, causes performance of operations comprising:
training, using a training data set comprising data signals associated with a set of data sources, a deterministic machine learning model to identify anomalous states of the data signals;
using the deterministic machine learning model, analyzing a target data set comprising a plurality of target data signals associated with the set of data sources to generate a plurality of operational results;
based on the plurality of operational results, receiving a root-cause request to identify a root cause of a particular operational result associated with the set of data sources;
generating an abductive model to identify correlations between a plurality of sub-sets of data sources, among the set of data sources, and respective operational results, among the plurality of operational results generated by the deterministic machine learning model, wherein generating the abductive model comprises:
varying, according to a first pre-defined pattern, input values to the deterministic machine learning model to identify a first sub-set of data sources that has a greater influence than a second sub-set of the data sources on determining which operational result is generated by the deterministic machine learning model; and
varying, according to a first pre-defined pattern, input values of third sub-sets of the first sub-set of data sources, without varying input values of the second sub-set of the data sources, to identify relative levels of influence of respective data sources on the deterministic machine learning model generating respective operational results; and
applying the abductive model to the particular operational result to identify a particular sub-set of the data sources, from among the plurality of sub-sets of data sources, that has a greatest influence on the deterministic machine learning model generating the particular operational result.