US 12,254,410 B2
Machine learning technique for automatic modeling of multiple-valued outputs
Tero Tapani Karras, Helsinki (FI); Timo Oskari Aila, Tuusula (FI); and Samuli Matias Laine, Vantaa (FI)
Assigned to NVIDIA Corporation, Santa Clara, CA (US)
Filed by NVIDIA Corporation, Santa Clara, CA (US)
Filed on May 5, 2023, as Appl. No. 18/312,692.
Application 18/312,692 is a continuation of application No. 15/826,430, filed on Nov. 29, 2017, granted, now 11,694,072.
Claims priority of provisional application 62/508,951, filed on May 19, 2017.
Prior Publication US 2024/0144001 A1, May 2, 2024
This patent is subject to a terminal disclaimer.
Int. Cl. G06N 3/08 (2023.01); G06F 18/21 (2023.01); G06F 18/2413 (2023.01); G06N 3/04 (2023.01); G06N 3/045 (2023.01); G06N 3/047 (2023.01); G06V 10/44 (2022.01); G06V 10/82 (2022.01); G06V 10/94 (2022.01); G06V 30/19 (2022.01); G10L 25/30 (2013.01)
CPC G06N 3/08 (2013.01) [G06F 18/217 (2023.01); G06F 18/24143 (2023.01); G06N 3/04 (2013.01); G06N 3/045 (2023.01); G06N 3/047 (2023.01); G06V 10/454 (2022.01); G06V 10/82 (2022.01); G06V 10/955 (2022.01); G06V 30/1916 (2022.01); G06V 30/19173 (2022.01); G10L 25/30 (2013.01)] 20 Claims
OG exemplary drawing
 
1. A method for augmenting input data for a model that implements a machine-learning algorithm, comprising:
augmenting an input vector with a first latent descriptor vector selected from a plurality of latent descriptor vectors stored in a database to produce an augmented input vector, wherein each latent descriptor vector stored in the database is mapped to a training identifier in a plurality of training identifiers; and
applying, by the model, a set of model parameters to the augmented input vector to generate an output vector, wherein the set of model parameters are trained by:
receiving a set of training data, wherein the set of training data includes a plurality of training input vectors, a plurality of training identifiers, and a plurality of desired output vectors;
providing a plurality of augmented input vectors as inputs to the model, wherein each augmented training input vector in the plurality of augmented input vectors comprises a training input vector in the plurality of training input vectors and a latent descriptor vector selected from the database using the training identifier that is associated with the training input vector;
applying, by the model, the set of model parameters to the plurality of augmented training input vectors to generate a set of output vectors; and
adjusting the set of model parameters and at least one latent descriptor vector stored in the database based on differences between each output vector in the set of output vectors and a corresponding desired output vector in the plurality of desired output vectors.