US 12,169,692 B2
Iterative training for multi-modal data in natural language processing
Adam Dancewicz, Warsaw (PL); Filip Gralinski, Warsaw (PL); and Lukasz Konrad Borchmann, Warsaw (PL)
Assigned to APPLICA SP. Z O.O., Warsaw (PL)
Filed by APPLICA SP. Z O.O., Warsaw (PL)
Filed on Jan. 31, 2024, as Appl. No. 18/428,859.
Application 18/428,859 is a continuation of application No. 18/127,458, filed on Mar. 28, 2023, granted, now 11,934,786.
Application 18/127,458 is a continuation of application No. 17/807,313, filed on Jun. 16, 2022, granted, now 11,620,451.
Application 17/807,313 is a continuation of application No. 17/651,313, filed on Feb. 16, 2022, granted, now 11,455,468.
Claims priority of provisional application 63/150,271, filed on Feb. 17, 2021.
Prior Publication US 2024/0211691 A1, Jun. 27, 2024
Int. Cl. G06F 17/00 (2019.01); G06F 40/106 (2020.01); G06F 40/295 (2020.01); G06F 40/30 (2020.01); G06N 3/08 (2023.01); G06T 11/60 (2006.01)
CPC G06F 40/295 (2020.01) [G06F 40/106 (2020.01); G06F 40/30 (2020.01); G06N 3/08 (2013.01); G06T 11/60 (2013.01)] 30 Claims
OG exemplary drawing
 
1. A system comprising:
one or more hardware processors of a machine; and
at least one memory storing instructions that, when executed by the one or more hardware processors, cause the machine to perform operations comprising:
performing a plurality of iterations to generate a Natural Language Processing (NLP) model, each iteration comprising:
receiving a plurality of real-world documents, the plurality of real-world documents including text data, layout data, and image data;
processing, by at least one or more hardware processors, the plurality of real-world documents to generate an initial prediction for data points within the plurality of real-world documents using a neural network;
validating the initial prediction by comparing extracted values corresponding with information present in the plurality of real-world documents and correcting discrepancies found based on the comparing;
evaluating a quality of the validated initial prediction; and
determining that the quality of the validated initial prediction satisfies a quality constraint; and
configuring the NLP model to process a new document to extract data points without validation.