Abstract: Each node in a subset of graph nodes has an associated label value indicating a characteristic of the corresponding item. Matrix data and graph label data are stored. The matrix data defines a matrix representing the graph. The graph label data defines a graph label vector indicating label values associated with nodes of the graph. For at least one set of nodes, test label data is generated defining a test label vector. A propagation function is defined, comprising a set of basis functions, having respective coefficients. The coefficients are calculated which minimize a difference function dependent on difference between the graph label vector and a result of applying the propagation function to the test label vector for said at least one set. New label values are calculated for nodes in K by applying the propagation function with the calculated coefficients to the graph label vector, thereby propagating labels.
Type:
Grant
Filed:
June 3, 2016
Date of Patent:
November 3, 2020
Assignees:
International Business Machines Corporation, UNIVERSITÈ LIBRE DE BRUXELLES BRUXELLES
Inventors:
Konstantinos Bekas, Robin Devooght, Peter Willem Jan Staar