US 11,921,697 B2
Methods and systems for detecting spurious data patterns
Louizos Alexandros Louizos, New York, NY (US); Ayaan Chaudhry, Toronto (CA); Gary Plunkett, Bellingham, WA (US); Oliver Clark, New York, NY (US); Cathy Ross, New York, NY (US); and R. Whitney Anderson, New York, NY (US)
Assigned to Fraud.net, Inc., New York, NY (US)
Filed by Fraud.net, Inc., New York, NY (US)
Filed on Nov. 20, 2020, as Appl. No. 17/100,195.
Claims priority of provisional application 62/939,236, filed on Nov. 22, 2019.
Prior Publication US 2021/0157786 A1, May 27, 2021
Int. Cl. G06F 16/23 (2019.01); G06F 16/28 (2019.01); G06N 3/04 (2023.01); G06N 3/08 (2023.01); G06Q 10/10 (2023.01); G06Q 30/018 (2023.01)
CPC G06F 16/2365 (2019.01) [G06F 16/285 (2019.01); G06N 3/04 (2013.01); G06N 3/08 (2013.01); G06Q 10/10 (2013.01); G06Q 30/0185 (2013.01)] 19 Claims
OG exemplary drawing
 
1. A method for detection and classification of data, the method comprising:
converting a set of data values comprising a multi-field record into a first graph representation of the multi-field record, with the first graph representation comprising nodes and edges, wherein converting the set of data values representative of the multi-field record comprises, for each of the data values comprising the multi-field record, transforming that data value from a 1-dimensional space into a respective multi-dimensional vector of a plurality of multi-dimensional vectors, wherein transforming that data value comprises using a corresponding trained multi-layer perceptron of a plurality of multi-layer perceptrons used to transform data values of the multi-field record, wherein each multi-dimensional vector of the plurality of multi-dimensional vectors corresponds to a respective one of the nodes of the first graph representation representing both feature information for respective data values of the multi-field record and a relationship of the respective one of the nodes to other of the nodes in the first graph representation;
applying a machine learning graph convolution process to the first graph representation of the multi-field record to generate a transformed graph representation for the multi-field record comprising a resultant transformed configuration of resultant transformed nodes and edges that organizes the resultant transformed nodes and edges according to relevance or a normality of the nodes and edges of the first graph representation of the multi-field record; and
determining, based on the resultant transformed configuration of the resultant transformed nodes and edges, a probability that the multi-field record is anomalous.