Abstract: The present disclosure relates to systems, methods, and non-transitory computer-readable media that generate a natural language model that provides user-entity differential privacy. For example, in one or more embodiments, the disclosed systems sample sensitive data points from a natural language dataset. Using the sampled sensitive data points, the disclosed systems determine gradient values corresponding to the natural language model. Further, the disclosed systems generate noise for the natural language model. The disclosed systems generate parameters for the natural language model using the gradient values and the noise, facilitating simultaneous protection of the users and sensitive entities associated with the natural language dataset. In some implementations, the disclosed systems generate the natural language model through an iterative process (e.g., by iteratively modifying the parameters).

Abstract: One example method involves operations for a processing device that include receiving, by a machine learning model trained to generate a search result, a search query for a text input. The machine learning model is trained by receiving pre-training data that includes multiple documents. Pre-training the machine learning model by generating, using an encoder, feature embeddings for each of the documents included in the pre-training data. The feature embeddings are generated by applying a masking function to visual and textual features in the documents. Training the machine learning model also includes generating, using the feature embeddings, output features for the documents by concatenating the feature embeddings and applying a non-linear mapping to the feature embeddings. Training the machine learning model further includes applying a linear classifier to the output features. Additionally, operations include generating, for display, a search result using the machine learning model based on the input.

Abstract: A device-management system performs processing, such as audio processing, in an instance of a virtual machine corresponding to a functionally limited (local) device. To register the local device, the device-management system receives a registration request that includes device information, encryption data, and an indication of an associated user account. The device-management system then sends this registration data to a service-provider system, which returns a shared encryption key. The device-management system and the local device may use this shared encryption key to securely communicate. The device-management system may de-allocate the instance upon detecting a period of inactivity of the local device and may re-allocate the instance when new activity is detected. The device-management system may further determine when and if audio data to be sent to the local device is encoded using a codec not implemented by the local device.

Abstract: A computer-implemented system includes a platform cloud server, and platform application software executable by the platform cloud server, the platform application software configured to close an access control gate to prevent a subject vehicle from exiting a secured vehicle parking facility at said access control gate unless an exit condition is satisfied, the exit condition including identity of the subject vehicle, receipt of an exit request to exit through the gate by touchless initiation, and actual receipt of payment by touchless initiation. A QR code is displayed to present the exit request, and a touchless vehicle detection system determines identity of the subject vehicle.

Abstract: Systems and methods for natural language processing are described. One or more embodiments of the present disclosure receive a document comprising a plurality of words organized into a plurality of sentences, the words comprising an event trigger word and an argument candidate word, generate word representation vectors for the words, generate a plurality of document structures including a semantic structure for the document based on the word representation vectors, a syntax structure representing dependency relationships between the words, and a discourse structure representing discourse information of the document based on the plurality of sentences, generate a relationship representation vector based on the document structures, and predict a relationship between the event trigger word and the argument candidate word based on the relationship representation vector.

Abstract: Certain embodiments involve transforming an electronic document into a tagged electronic document. For instance, an electronic document processing application generates a tagged electronic document from an input electronic document. The electronic document processing application accesses one or more feature maps that identify, via a set of object-recognition rules, identified objects in the electronic document. The electronic document processing application also obtains a heat map of the electronic document that represents attributes in a pixel-wise manner. The electronic document processing application computes a tag by applying a fusion deep learning model to the one or more feature maps and the heat map. The electronic document processing application generates the tagged electronic document by applying the tag to the electronic document.

Abstract: A device-management system performs processing, such as audio processing, in an instance of a virtual machine corresponding to a functionally limited (local) device. To register the user device, the device-management system receives a registration request that includes device information, encryption data, and an indication of an associated user account. The device-management system then sends this registration data to a service-provider system, which returns a shared encryption key. The device-management system and the user device may use this shared encryption key to securely communicate. The device-management system may de-allocate the instance upon detecting a period of inactivity of the user device and may re-allocate the instance when new activity is detected. The device-management system may further determine when and if audio data to be sent to the user device is encoded using a codec not implemented by the user device.

Abstract: Certain embodiments involve using a machine-learning tool to generate metadata identifying segments and topics for text within a document. For instance, in some embodiments, a text processing system obtains input text and applies a segmentation-and-labeling model to the input text. The segmentation-and-labeling model is trained to generate a predicted segment for the input text using a segmentation network. The segmentation-and-labeling model is also trained to generate a topic for the predicted segment using a pooling network of the model to the predicted segment. The output of the model is usable for generating metadata identifying the predicted segment and the associated topic.

Abstract: Techniques are provided for generating a digital image of simulated handwriting using an encoder-decoder neural network trained on images of natural handwriting samples. The simulated handwriting image can be generated based on a style of a handwriting sample and a variable length coded text input. The style represents visually distinctive characteristics of the handwriting sample, such as the shape, size, slope, and spacing of the letters, characters, or other markings in the handwriting sample. The resulting simulated handwriting image can include the text input rendered in the style of the handwriting sample. The distinctive visual appearance of the letters or words in the simulated handwriting image mimics the visual appearance of the letters or words in the handwriting sample image, whether the letters or words in the simulated handwriting image are the same as in the handwriting sample image or different from those in the handwriting sample image.

Abstract: The present disclosure relates to systems, non-transitory computer-readable media, and methods for classifying an input image utilizing a classification model conditioned by a generative model and/or self-supervision. For example, the disclosed systems can utilize a generative model to generate a reconstructed image from an input image to be classified. In turn, the disclosed systems can combine the reconstructed image with the input image itself. Using the combination of the input image and the reconstructed image, the disclosed systems utilize a classification model to determine a classification for the input image. Furthermore, the disclosed systems can employ self-supervised learning to cause the classification model to learn discriminative features for better classifying images of both known classes and open-set categories.

Abstract: This disclosure describes methods, non-transitory computer readable storage media, and systems that extract a definition for a term from a source document by utilizing a single machine-learning framework to classify a word sequence from the source document as including a term definition and to label words from the word sequence. To illustrate, the disclosed system can receive a source document including a word sequence arranged in one or more sentences. The disclosed systems can utilize a machine-learning model to classify the word sequence as comprising a definition for a term and generate labels for the words from the word sequence corresponding to the term and the definition. Based on classifying the word sequence and the generated labels, the disclosed system can extract the definition for the term from the source document.

Abstract: Techniques are provided for generating a digital image of simulated handwriting using an encoder-decoder neural network trained on images of natural handwriting samples. The simulated handwriting image can be generated based on a style of a handwriting sample and a variable length coded text input. The style represents visually distinctive characteristics of the handwriting sample, such as the shape, size, slope, and spacing of the letters, characters, or other markings in the handwriting sample. The resulting simulated handwriting image can include the text input rendered in the style of the handwriting sample. The distinctive visual appearance of the letters or words in the simulated handwriting image mimics the visual appearance of the letters or words in the handwriting sample image, whether the letters or words in the simulated handwriting image are the same as in the handwriting sample image or different from those in the handwriting sample image.

Abstract: Techniques are provided for generating a digital image of simulated handwriting using an encoder-decoder neural network trained on images of natural handwriting samples. The simulated handwriting image can be generated based on a style of a handwriting sample and a variable length coded text input. The style represents visually distinctive characteristics of the handwriting sample, such as the shape, size, slope, and spacing of the letters, characters, or other markings in the handwriting sample. The resulting simulated handwriting image can include the text input rendered in the style of the handwriting sample. The distinctive visual appearance of the letters or words in the simulated handwriting image mimics the visual appearance of the letters or words in the handwriting sample image, whether the letters or words in the simulated handwriting image are the same as in the handwriting sample image or different from those in the handwriting sample image.

Abstract: Certain embodiments involve transforming an electronic document into a tagged electronic document. For instance, an electronic document processing application generates a tagged electronic document from an input electronic document. The electronic document processing application accesses one or more feature maps that identify, via a set of object-recognition rules, identified objects in the electronic document. The electronic document processing application also obtains a heat map of the electronic document that represents attributes in a pixel-wise manner. The electronic document processing application computes a tag by applying a fusion deep learning model to the one or more feature maps and the heat map. The electronic document processing application generates the tagged electronic document by applying the tag to the electronic document.

Abstract: In some embodiments, a computing system computes tags for an electronic document. The computing system identifies sets of objects for the electronic document by applying a set of object-recognition rules to the electronic document, with each object-recognition rule generating a set of identified objects. The computing system generates feature maps that represent a set of identified objects. The computing system generates a heat map that identifies attributes of the electronic document including object candidates of the electronic document by applying a page-segmentation machine-learning model to the electronic document. The computing system computes a tag by applying a fusion deep learning module to the feature map and the heat map to correlate a document object identified by the feature map with an attribute of the electronic document identified by the heat map. The computing system generates the tagged electronic document by applying the tag to the electronic document.

Abstract: A device-management system performs audio processing, such as acoustic echo cancellation or beamforming, in a computing-resource allocation corresponding to a functionally limited device. The device-management system may be a locally-connected network device that is in communication with one or more user devices; the device-management system may also or instead be a remote device that communicates with the user devices using the locally-connected network device. The device-management system may receive audio data from one or more microphones of one or more user devices. To perform acoustic echo cancellation, the device-management system may receive and process time data corresponding to a time of output of audio by the user device.

Abstract: In some embodiments, a computing system computes tags for an electronic document. The computing system identifies sets of objects for the electronic document by applying a set of object-recognition rules to the electronic document, with each object-recognition rule generating a set of identified objects. The computing system generates feature maps that represent a set of identified objects. The computing system generates a heat map that identifies attributes of the electronic document including object candidates of the electronic document by applying a page-segmentation machine-learning model to the electronic document. The computing system computes a tag by applying a fusion deep learning module to the feature map and the heat map to correlate a document object identified by the feature map with an attribute of the electronic document identified by the heat map. The computing system generates the tagged electronic document by applying the tag to the electronic document.

Abstract: In an embodiment, a computer system comprises one or more computer processors configured with a message transfer application; a message transfer/vision processing (MT/VP) interface coupled to the one or more computer processors and interposed between the message transfer application and a vision processing computer, wherein the MT/VP interface performs operations comprising: extracting risk indicator data from a message that is in transit to a recipient computer on a computer network; in response to the risk indicator data matching a message risk criterion, transmitting an image address for an image of interest coupled to the message or the image of interest to the vision processing computer; receiving, from the vision processing computer, a label that semantically describes visual content of the image of interest; using the label, querying a set of correlation data to determine a reference address that is associated with the label; in response to the image address matching the reference address, transmitting

Abstract: Determining whether a subject tax return is fraudulent includes extracting from the subject tax return information and identifying one or more subject nodes based on the extracted information. Separately, a plurality of external nodes is generated based upon previously filed tax returns. At least a portion of the plurality of external nodes is fraud-indicative nodes. The subject nodes are compared to the external nodes to identify shared relationships of related information, such as a tax return related to an external node having the same bank account information as the subject tax return related to the subject node. Based upon shared information, links are determined to indicate whether the subject node is indicative of fraud.

Abstract: Systems and methods including one or more processing modules and one or more non-transitory storage modules storing computing instructions configured to run on the one or more processing modules and perform acts of determining a product shipping cost for a product and coordinating a display on the electronic device of the user of the product shipping cost of a carrier shipping cost that is less than the maximum shipping cost for the product. The maximum shipping cost for the product can be determined by determining carriers comprising a transit time for shipping the product that is less than or equal to a shipping time requirement of a service level agreement with the user, determining a baseline cost for shipping the product, retrieving a fixed threshold cost for the product, and combining the fixed threshold cost and the baseline cost to determine a maximum shipping cost for the product.