US 11,741,693 B2
System and method for semi-supervised conditional generative modeling using adversarial networks
Sricharan Kallur Palli Kumar, Mountain View, CA (US); Raja Bala, Pittsford, NY (US); Jin Sun, Redwood City, MD (US); Hui Ding, College Park, MD (US); and Matthew A. Shreve, Mountain View, CA (US)
Assigned to Palo Alto Research Center Incorporated, Palo Alto, CA (US)
Filed by Palo Alto Research Center Incorporated, Palo Alto, CA (US)
Filed on Nov. 29, 2017, as Appl. No. 15/826,613.
Claims priority of provisional application 62/586,786, filed on Nov. 15, 2017.
Prior Publication US 2019/0147333 A1, May 16, 2019
Int. Cl. G06V 10/82 (2022.01); G06N 3/08 (2023.01); G06F 18/2413 (2023.01); G06V 10/764 (2022.01); G06V 10/44 (2022.01)
CPC G06V 10/82 (2022.01) [G06F 18/2413 (2023.01); G06N 3/08 (2013.01); G06V 10/451 (2022.01); G06V 10/764 (2022.01)] 16 Claims
OG exemplary drawing
 
1. A computer-implemented method for generating synthetic data objects using a semi-supervised generative adversarial network, the method comprising:
synthesizing, by a generator module, a data object xG derived from a noise vector z and an attribute label y,
wherein the semi-supervised generative adversarial network comprises the generator module, an unsupervised discriminator module, and a supervised discriminator module;
passing, to the unsupervised discriminator module, the data object xG and a set of training objects xT and xU which are obtained from a training data set,
wherein the training data set includes the xU objects which do not have a corresponding attribute label and further includes the xT objects that do have corresponding attribute labels yT;
calculating, by the unsupervised discriminator module, a value indicating a probability that the data object xG is real;
producing, by the unsupervised discriminator module, a first latent feature representation h(xG) of the data object xG and a second latent feature representation h(xT) of the data object xT;
passing the first and second latent feature representations h(xG) and h(xT) to the supervised discriminator module;
passing the attribute label y and an attribute label yT corresponding to the data object xT to the supervised discriminator module;
receiving, by the supervised discriminator module as input, a first pair comprising the first latent feature representation h(xG) produced by the unsupervised discriminator module and the attribute label y and a second pair comprising the second latent feature representation h(xT) produced by the unsupervised discriminator module and the corresponding yT attribute;
calculating, by the supervised discriminator module based on at least the first pair and the second pair, a value indicating a probability that the attribute label y given the data object xG is real; and
performing the aforementioned steps iteratively until the generator module produces data objects with a given attribute label which the unsupervised and supervised discriminator modules can no longer identify as fake.