Abstract: Systems, methods, and non-transitory computer-readable media can identify a set of accounts, each account of the set of accounts having a number of followers. The set of accounts are grouped into a plurality of groups based on number of followers, wherein each group is associated with a value score. A machine learning model is trained using a set of training data comprising account recommendation conversion information, wherein the account recommendation conversion information comprises a plurality of successful account recommendations, and each successful account recommendation is assigned a weight based on the value scores associated with the plurality of groups. One or more accounts of the set of accounts are selected to present as account recommendations based on the machine learning model.

Abstract: A content analyzer determines whether various types of modification have been made to images. The content analyzer computes JPEG ghosts from the images that are concatenated with the image channels to generate a feature vector. The feature vector is provided as input to a neural network that determines whether the types of modification have been made to the image. The neural network may include a constrained convolution layer and several unconstrained convolution layers. An image fake model may also be applied to determine whether the image was generated using a computer model or algorithm.

Abstract: In one embodiment, a method includes accessing a first machine-learning model trained to generate a feature representation of an input data, a second machine-learning model trained to generate a desired result based on the feature representation, and a third machine-learning model trained to generate an undesired result based on the feature representation, and training a fourth machine-learning model by generating a secured feature representation by processing a first output of the first machine-learning model using the fourth machine-learning model, generating a second output and a third output by processing the secured feature representation using, respectively, the second and third machine-learning models, and updating the fourth machine-learning model according to an optimization function configured to optimize a correctness of the second output and an incorrectness of the third output.

Abstract: Systems, methods, and non-transitory computer-readable media can be configured to train a featurizer based at least in part on a set of training data. The featurizer can be applied to at least one input to generate at least one tensor. The at least one tensor obfuscates or excludes at least one feature in the at least one input.

Abstract: A content analyzer determines whether various types of modification have been made to images. The content analyzer computes JPEG ghosts from the images that are concatenated with the image channels to generate a feature vector. The feature vector is provided as input to a neural network that determines whether the types of modification have been made to the image. The neural network may include a constrained convolution layer and several unconstrained convolution layers. An image fake model may also be applied to determine whether the image was generated using a computer model or algorithm.

Abstract: In one embodiment, a method includes generating depth map for a reference image and generating a three-dimensional (3D) model for a plurality of objects in the reference image based on the depth map. The method additionally includes determining, out of the objects in the 3D model, a background object having a boundary adjacent to a foreground object. The method also includes determining that at least a portion of a surface of the background object is hidden by the foreground object and extending, in the 3D model, the surface of the background object to include the portion hidden by the foreground object. The method further includes in-paint pixels of the extended surface of the background object with pixels that approximate the portion of the surface of the background object hidden by the foreground object.

Abstract: In one embodiment, a computing system may access a training image and a reference image of a person and an incomplete image. A generate may generate an in-painted image based on the incomplete image, and a discriminator may be used to determine whether each of the in-painted image, the training image, and the reference image is likely generated by the generator. The system may compute losses based on the determinations and update the discriminator accordingly. Using the updated discriminator, the system may determine whether a second in-painted image generated by the generator is likely generated by the generator. The system may compute a loss based on the determination and update the generator accordingly. Once training is complete, the generator may be used to generate a modified version of a given image, such as making the eyes of a person appear open even if they were closed in the input image.

Abstract: In one embodiment, a computing system may access a training image and a reference image of a person and an incomplete image. A generate may generate an in-painted image based on the incomplete image, and a discriminator may be used to determine whether each of the in-painted image, the training image, and the reference image is likely generated by the generator. The system may compute losses based on the determinations and update the discriminator accordingly. Using the updated discriminator, the system may determine whether a second in-painted image generated by the generator is likely generated by the generator. The system may compute a loss based on the determination and update the generator accordingly. Once training is complete, the generator may be used to generate a modified version of a given image, such as making the eyes of a person appear open even if they were closed in the input image.