Abstract: A summary generation and summary selection system is disclosed that is capable of automatically evaluating multiple summaries generated for content and selecting a single summary that is deemed to be the “best” among the multiple generated summaries. The system includes capabilities to use multiple different selection techniques to select the best summary from multiple generated summaries. A first selection technique involves identifying entities and entity relationships from the content to be summarized and selecting a summary from multiple summaries generated for the content based on the entities and entity relationships identified in the content. A second selection technique involves determining a set of questions that are answered by each summary. The technique then selects a summary based upon the set of questions answered by each summary. The system then outputs the selected summary as the summary for the content.

Abstract: The present embodiments relate to a language identification system for predicting a language and text content of text lines in an image-based document. The language identification system uses a trainable neural network model that integrates multiple neural network models in a single unified end-to-end trainable architecture. A CNN and an RNN of the model can process text lines and derive visual and contextual features of the text lines. The derived features can be used to predict a language and text content for the text line. The CNN and the RNN can be jointly trained by determining losses based on the predicted language and content and corresponding language labels and text labels for each text line.

Abstract: Techniques for layout-aware multi-modal networks for document understanding are provided. In one technique, word data representations that were generated based on words that were extracted from an image of a document are identified. Based on the image, table features of one or more tables in the document are determined. One or more table data representations that were generated based on the table features are identified. The word data representations and the one or more table data representations are input into a machine-learned model to generate a document data representation for the document. A task is performed based on the document data representation. In a related technique, instead of the one or more table data representations, one or more layout data representations that were generated based on a set of layout features, of the document, that was determined based on the image are identified and input into the machine-learned model.

Abstract: Techniques are disclosed for optical character recognition of extensible markup language content. A method can include a system generating a first training data comprising extensible markup language (XML) content, the first training data comprising a first plurality of training instances, each training instance including a respective image comprising XML content and annotation information for the respective image. The system can train a plurality of machine learning models using the first training data to generate a plurality of trained machine learning models, to perform image-based XML content extraction. The system can generate a plurality of trained machine learning models based at least in part on the training.

Abstract: An efficient and robust high-speed neural networks for cell image classification is described herein. The neural networks for cell image classification utilize a difference between soft-max scores to determine if a cell is ambiguous or not. If the cell is ambiguous, then the class is classified in a pseudo class, and if the cell is not ambiguous, the cell is classified in the class corresponding to the highest class score. The neural networks for cell image classification enable a high speed, high accuracy and high recall implementation.

Abstract: A method for super-resolution of block-compressed texture is provided. The method includes receiving a first texture block of a first block size. Based on application of a first block compression (BC) scheme on the received first texture block, coded-texture values are generated in a compressed domain. Further, a first machine learning model is applied on the generated coded-texture values to generate super-resolution coded-texture values in the compressed domain. The generated super-resolution coded-texture values are processed to generate a second texture block of a second block size. The second block size is greater than the first block size.

Abstract: An image-based classification workflow uses unsupervised clustering to help a user identify subpopulations of interest for sorting. Labeled cell images are used to fine-tune the supervised classification network for a specific experiment. The workflow allows the user to select the populations to sort in the same manner for a variety of applications. The supervised classification network is very fast, allowing it to make real-time sort decisions as the cell travels through a device. The workflow is more automated and has fewer user steps, which improves the ease of use. The workflow uses machine learning to avoid human error and bias from manual gating. The workflow does not require the user to be an expert in image processing, thus increasing the ease of use.

Abstract: Techniques are described for automatically, and substantially without human intervention, generating training data where the training data includes a set of training images containing text content and associated label data. Both the training images and the associated label data are automatically generated. The label data that is automatically generated for a training image includes one or more labels identifying locations of one or more text portions within the training image, and for each text portion, a label indicative of the text content in the text portion. By automating both the generation of training images and the generation of associated label data, the techniques described herein are very scalable and repeatable and can be used to generate large amounts of training data, which in turn enables building more reliable and accurate language models.

Abstract: The present embodiments relate to a language identification system for predicting a language and text content of text lines in an image-based document. The language identification system uses a trainable neural network model that integrates multiple neural network models in a single unified end-to-end trainable architecture. A CNN and an RNN of the model can process text lines and derive visual and contextual features of the text lines. The derived features can be used to predict a language and text content for the text line. The CNN and the RNN can be jointly trained by determining losses based on the predicted language and content and corresponding language labels and text labels for each text line.

Abstract: The present embodiments relate to identifying a native language of text included in an image-based document. A cloud infrastructure node (e.g., one or more interconnected computing devices implementing a cloud infrastructure) can utilize one or more deep learning models to identify a language of an image-based document (e.g., a scanned document) that is formed of pixels. The cloud infrastructure node can detect text lines that are bounded by bounding boxes in the document, determine a primary script classification of the text in the document, and derive a primary language for the document. Various document management tasks can be performed responsive to determining the language, such as perform optical character recognition (OCR) or derive insights into the text.

Abstract: An image-based classification workflow uses unsupervised clustering to help a user identify subpopulations of interest for sorting. Labeled cell images are used to fine-tune the supervised classification network for a specific experiment. The workflow allows the user to select the populations to sort in the same manner for a variety of applications. The supervised classification network is very fast, allowing it to make real-time sort decisions as the cell travels through a device. The workflow is more automated and has fewer user steps, which improves the ease of use. The workflow uses machine learning to avoid human error and bias from manual gating. The workflow does not require the user to be an expert in image processing, thus increasing the ease of use.

Abstract: An efficient and robust high-speed neural networks for cell image classification is described herein. The neural networks for cell image classification utilize a difference between soft-max scores to determine if a cell is ambiguous or not. If the cell is ambiguous, then the class is classified in a pseudo class, and if the cell is not ambiguous, the cell is classified in the class corresponding to the highest class score. The neural networks for cell image classification enable a high speed, high accuracy and high recall implementation.

Abstract: A method for super-resolution of block-compressed texture is provided. The method includes receiving a first texture block of a first block size. Based on application of a first block compression (BC) scheme on the received first texture block, coded-texture values are generated in a compressed domain. Further, a first machine learning model is applied on the generated coded-texture values to generate super-resolution coded-texture values in the compressed domain. The generated super-resolution coded-texture values are processed to generate a second texture block of a second block size. The second block size is greater than the first block size.