Abstract: Machine learning techniques are disclosed for rebuilding transactions to predict cash position. In one aspect a method includes obtaining data for an original transaction, classifying the original transaction into a class of multiple classes based on the data, predicting first tranche delay days for the original transaction based on the class and the data, predicting a tranche count for the original transaction based on the class and the data, predicting a tranche interval for the original transaction based on the class and the data; and rebuilding the original transaction as one or more future transactions based on the class, the first tranche delay days, the tranche count, and tranche interval. Each of the one or more future transactions comprise an updated amount of the original transaction, an updated date upon which the original transaction is anticipated, or both.

Abstract: A solution for managing a video conference is provided. Multiple segmentation areas or shapes are stored and made transparent. The background is recorded continuously and simultaneously in a memory presented to a user for a second participant when they are participating in a video conference with a first participant. The segmentation or transparency of objects obstructing information in view of the video conference is selected based on attributes of the first or second participant, or attributes of the video conference. The segmentation may be utilized to provide a desired view, message, or information to individuals participating in, watching, or interacting with a first participant via video conferencing. The background includes static images, a live video feed, a recorded video feed, a static webpage, a non-static webpage, on any surface without obstruction when a first participant is in front of the surface sharing information due to segmentation/transparency of the first participant.

Abstract: Machine learning techniques are disclosed for rebuilding transactions to predict cash position. In one aspect a method includes obtaining data for an original transaction, classifying the original transaction into a class of multiple classes based on the data, predicting first tranche delay days for the original transaction based on the class and the data, predicting a tranche count for the original transaction based on the class and the data, predicting a tranche interval for the original transaction based on the class and the data; and rebuilding the original transaction as one or more future transactions based on the class, the first tranche delay days, the tranche count, and tranche interval. Each of the one or more future transactions comprise an updated amount of the original transaction, an updated date upon which the original transaction is anticipated, or both.

Abstract: Multiple users communicate over a network via client devices that include one or more cameras and a display to enable video messaging. At least one of the client devices modifies regions of video data captured by the client device's camera to more prominently identify the people within the video data. To identify a person, the client device disambiguates between actual people and static objects that may appear like people. The client device uses pose models to identify bounding boxes and applies a motion model to determine if a bounding box may include a person based on an amount of movement within the bounding box. If a threshold amount of movement is detected in a bounding box, the client device obtains a higher resolution portion of the scene including the bounding box and classifies whether the bounding box contains a person based on movement within the higher resolution video.

Abstract: Approaches introduce a pre-processing and post-processing framework to a neural network-based approach to identify items represented in an image. For example, a classifier that is trained on several categories can be provided. An image that includes a representation of an item of interest is obtained. Rotated versions of the image are generated and each of a subset of the rotated images is analyzed to determine a probability that a respective image includes an instance of a particular category. The probabilities can be used to determine a probability distribution of output category data, and the data can be analyzed to select an image of the rotated versions of the image. Thereafter, a categorization tree can then be utilized, whereby for the item of interest represented the image, the category of the item can be determined. The determined category can be provided to an item retrieval algorithm to determine primary content for the item of interest.

Abstract: Approaches introduce a pre-processing and post-processing framework to a neural network-based approach to identify items represented in an image. For example, a classifier that is trained on several categories can be provided. An image that includes a representation of an item of interest is obtained. Rotated versions of the image are generated and each of a subset of the rotated images is analyzed to determine a probability that a respective image includes an instance of a particular category. The probabilities can be used to determine a probability distribution of output category data, and the data can be analyzed to select an image of the rotated versions of the image. Thereafter, a categorization tree can then be utilized, whereby for the item of interest represented the image, the category of the item can be determined. The determined category can be provided to an item retrieval algorithm to determine primary content for the item of interest.