Abstract: A computing device may receive a set of user documents. Data may be extracted from the documents to generate a first graph data structure with one or more initial graphs containing key-value pairs. A model may be trained on the first graph data structure to classify the pairs. Until a set of evaluation metrics for the model exceeds a set of deployment thresholds: generating, a set of evaluation metrics may be generated for the model. The set of evaluation metrics may be compared to the set of deployment thresholds. In response to a determination that the set of evaluation metrics are below the set of deployment thresholds: one or more new graphs may be generated from the one or more initial graphs in the first graph data structure to produce a second graph data structure. The first and second graph can be used to train the model.

Abstract: Continual learning techniques are described for extending the capabilities of a base model, which is trained to predict a set of existing or base classes, to generate a target model that is capable of making predictions for both the existing or base classes and additionally for making predictions for new or custom classes. The techniques described herein enable the target model to be trained such that the model can make predictions involving both base classes and custom classes with high levels of accuracy.

Abstract: A computing device may access visually rich documents comprising an image and metadata. A graph, based on the image or metadata, can be generated for a visually rich document. The graph's nodes can correspond to words from the visually rich document. Features for nodes can be determined by the device. The device may generate model labeled graphs by assigning a pseudo-label to nodes using a pretrained model. The device may generate a plurality of graph labeled graphs by assigning a pseudo-label to nodes by matching a first node from a first graph to at least a second node from a second graph. The device may generate a plurality of updated graphs by cross referencing labels from the model labeled graphs and the graph labeled graphs. Until a change in labels is below a threshold, a model can be trained to perform key-value extraction using the updated graphs.

Abstract: Embodiments described herein are directed towards a synthetic document generation pipeline for training artificial intelligence models. One embodiment includes a method including a device that receives an instruction to generate a document to be used as a training instance for a first machine learning model, the instruction including an element configuration, a document class configuration, a format configuration, an augmentation configuration, and data bias and fairness. The device can receive an element from an interface based at least in part on the element configuration, the element can simulate a real-world image, real-world text, or real-world machine-readable visual code. The device can generate metadata describe a layout for the element on the document based on the document class configuration. The device can generate the document by arranging the element on the document based on the metadata, wherein the document is generated in a format based on the format configuration.

Abstract: A computing device may access visually rich documents comprising an image and metadata. A graph, based on the image or metadata, can be generated for a visually rich document. The graph's nodes can correspond to words from the visually rich document. Features for nodes can be determined by the device. The device may generate model labeled graphs by assigning a pseudo-label to nodes using a pretrained model. The device may generate a plurality of graph labeled graphs by assigning a pseudo-label to nodes by matching a first node from a first graph to at least a second node from a second graph. The device may generate a plurality of updated graphs by cross referencing labels from the model labeled graphs and the graph labeled graphs. Until a change in labels is below a threshold, a model can be trained to perform key-value extraction using the updated graphs.

Abstract: A computing device may access visually rich documents comprising an image and metadata. A graph, based on the image or metadata, can be generated for a visually rich document. The graph's nodes can correspond to words from the visually rich document. Features for nodes can be determined by the device. The device may generate model labeled graphs by assigning a pseudo-label to nodes using a pretrained model. The device may generate a plurality of graph labeled graphs by assigning a pseudo-label to nodes by matching a first node from a first graph to at least a second node from a second graph. The device may generate a plurality of updated graphs by cross referencing labels from the model labeled graphs and the graph labeled graphs. Until a change in labels is below a threshold, a model can be trained to perform key-value extraction using the updated graphs.

Abstract: A computing device may receive a set of user documents. Data may be extracted from the documents to generate a first graph data structure with one or more initial graphs containing key-value pairs. A model may be trained on the first graph data structure to classify the pairs. Until a set of evaluation metrics for the model exceeds a set of deployment thresholds: generating, a set of evaluation metrics may be generated for the model. The set of evaluation metrics may be compared to the set of deployment thresholds. In response to a determination that the set of evaluation metrics are below the set of deployment thresholds: one or more new graphs may be generated from the one or more initial graphs in the first graph data structure to produce a second graph data structure. The first and second graph can be used to train the model.

Abstract: A video preview creation system creates portrait-mode video previews from landscape-mode video content by analyzing video frames to find candidate segments for the video preview. The candidate segments are filtered to find frames that are desirable using quality-based rules. Filtered segments are then smart-cropped and stitched together to create the portrait video preview.

Abstract: Approaches, techniques, and mechanisms are disclosed for generating subscriptions. According to one embodiment, one or more local features of an input request for service subscription are generated based at least in part on one or more messages originated from a client device that represent the input request. One or more global features of a population of input requests originated from a population of client devices are determined based at least in part on a population of input requests. One or more mapped global features of the input request are generated from the one or more global features via one or more mapping functions. One or more machine learning (ML) based prediction models are applied to the one or more local features and the one or more mapped global features of the input request to compute a fraud score for the input request. The fraud score for the input request is used to determine whether the input request for service subscription is to be accepted.

Abstract: The present disclosure proposes an image processing method according to an embodiment. The image processing method includes the operations of obtaining a color component value from an input image, determining a dynamic range of the obtained color component value, determining a scaling factor for converting the dynamic range of the color component value, based on the determined dynamic range of the color component value and a permissible range of the color component value, and scaling the color component value, based on the determined scaling factor.

Abstract: Approaches, techniques, and mechanisms are disclosed for generating thumbnails. According to one embodiment, a subset of images each depicting character face(s) is identified from a collection of images. An unsupervised learning method is applied to automatically cluster the subset of images into image clusters. Top image clusters are selected from the image clusters based at least in part on weighted scores of images clustered within the image clusters. Thumbnail(s) are generated from images in the top image clusters.

Abstract: Approaches, techniques, and mechanisms are disclosed for generating thumbnails. According to one embodiment, a subset of images each depicting character face(s) is identified from a collection of images. An unsupervised learning method is applied to automatically cluster the subset of images into image clusters. Top image clusters are selected from the image clusters based at least in part on weighted scores of images clustered within the image clusters. Thumbnail(s) are generated from images in the top image clusters.

Abstract: The present disclosure proposes an image processing method according to an embodiment. The image processing method includes the operations of obtaining a color component value from an input image, determining a dynamic range of the obtained color component value, determining a scaling factor for converting the dynamic range of the color component value, based on the determined dynamic range of the color component value and a permissible range of the color component value, and scaling the color component value, based on the determined scaling factor.

Abstract: Approaches, techniques, and mechanisms are disclosed for generating thumbnails. According to one embodiment, a subset of images each depicting character face(s) is identified from a collection of images. An unsupervised learning method is applied to automatically cluster the subset of images into image clusters. Top image clusters are selected from the image clusters based at least in part on weighted scores of images clustered within the image clusters. Thumbnail(s) are generated from images in the top image clusters.