US 11,853,395 B2
Augmentation loss function for image classification
Yoel Shoshan, Haifa (IL); and Vadim Ratner, Haifa (IL)
Assigned to International Business Machines Corporation, Armonk, NY (US)
Filed by International Business Machines Corporation, Armonk, NY (US)
Filed on Jul. 2, 2020, as Appl. No. 16/919,130.
Prior Publication US 2022/0004823 A1, Jan. 6, 2022
Int. Cl. G06F 18/241 (2023.01); G06N 3/084 (2023.01); G06F 18/211 (2023.01)
CPC G06F 18/241 (2023.01) [G06F 18/211 (2023.01); G06N 3/084 (2013.01)] 20 Claims
OG exemplary drawing
 
1. A computer-implemented method comprising:
inputting corresponding pairs of a plurality of training images to an image classifier, wherein respective pairs of the corresponding pairs comprise at least two images having a same classification and different augmentations; and
training an artificial neural network of the image classifier to classify the plurality of training images using an augmentation loss function, wherein the augmentation loss function reduces differences in model outputs between the corresponding pairs of the plurality of training images, and wherein a difference in the model outputs for a first pair of the corresponding pairs is associated with a coefficient, wherein the coefficient is relatively larger when the different augmentations comprise a single augmentation, and wherein the coefficient is relatively smaller when the different augmentations comprise a combination of multiple augmentations.