US 12,093,848 B2
Learning parameters of Bayesian network using uncertain evidence
Eliezer Segev Wasserkrug, Haifa (IL); and Radu Marinescu, Dublin (IE)
Assigned to International Business machines Corporation, Armonk, NY (US)
Filed by International Business Machines Corporation, Armonk, NY (US)
Filed on Dec. 1, 2020, as Appl. No. 17/107,984.
Prior Publication US 2022/0172091 A1, Jun. 2, 2022
Int. Cl. G06N 7/02 (2006.01); G06N 7/01 (2023.01); G06N 20/00 (2019.01)
CPC G06N 7/02 (2013.01) [G06N 7/01 (2023.01); G06N 20/00 (2019.01)] 21 Claims
OG exemplary drawing
 
1. A method for learning parameters of Bayesian network from uncertain evidence, comprising:
receiving an input comprising graph representation and at least one sample of a Bayesian network, wherein the graph comprising a plurality of nodes representing random variables and a plurality of directed edges between pairs of nodes from the plurality of nodes representing conditional dependencies, wherein each of the at least one sample comprising for each node of the plurality of nodes a value selected from the group consisting of: a known value; an unknown value; and an uncertain value comprising an observed value and associated confidence level between 0 and 1; wherein the at least one sample comprising one or more uncertain values in one or more samples;
applying on the input Bayesian network learning comprising:
inferring probabilities of random variables represented by the plurality of nodes, by applying on the input Bayesian network uncertain inference comprising performing inference in the Bayesian network from the at least one sample comprising the one or more uncertain values in the one or more samples; and
calculating estimates of conditional probability tables of the Bayesian network using the probabilities inferred; and
performing a machine learning based task using the estimates of conditional probability tables calculated.