Abstract: Systems and techniques are provided for self-healing at the edge. A plurality of faults can be detected for an edge device based on monitoring log information of the edge device, with remediation information indicative of remediation actions performed for individual faults or combinations of faults included in the plurality of faults. A hierarchical fault tree data structure can be generated mapping between the plurality of faults and the remediation information, each fault comprising root or child node of the fault tree, and each remediation action comprising a leaf node of the fault tree. An indication of one or more faults can be provided to a self-healing machine-learning or artificial intelligence engine configured to generate a corresponding fault remediation action based on traversing the hierarchical fault tree according to the indicated one or more faults. The fault remediation action can be output for self-healing of the indicated one or more faults.

Abstract: Systems and techniques are provided for multi-WAN routing and management at the edge. A multi-WAN network device can configure first and second backhaul connections between a local edge network and a remote WAN, the first and second backhaul connections implemented using first and second communication modalities different from one another and included in a plurality of communication modalities of the multi-WAN network device. A first containerized application on the multi-WAN network device can obtain telemetry or monitoring information generated by a satellite internet constellation terminal in the local edge network, and can transmit the information to a cloud management entity using one or more of the first or second backhaul connections of the multi-WAN network device. The multi-WAN network device can implement a network configuration adjustment for one or more of the first or second backhaul connection based at least in part on the telemetry information or monitoring information.

Abstract: Systems and techniques are provided for resiliency and redundancy for provisioning and/or configuring an edge compute unit. Configuration information can be obtained for provisioning an edge device with a plurality of nodes each associated with a respective rack of a plurality of racks. A first subset of the plurality of nodes can be provisioned, based on the configuration information, as a management cluster for workloads deployed to the edge device, the management cluster provisioned to include multiple redundant management control plane nodes distributed across different racks of the plurality of racks. A workload cluster can be provisioned on a remaining portion of the plurality of nodes, the workload cluster provisioned to include: multiple redundant workload control plane nodes distributed across different racks of the plurality of racks, and a respective plurality of worker nodes provisioned on each rack of the plurality of racks.

Abstract: Systems and techniques are disclosed for machine learning based improvements in estimation techniques. An example method includes obtaining a seed specifying a portion of an action for evaluation with respect to a networked computing environment. A description of the action is completed by providing a first recurrent neural network (RNN) with the seed to generate one or more additional words, with the action represented by the description not being enabled yet in the networked computing environment. An estimated down-stream impact (DSI) associated with the action is determined based on a second RNN, with the estimated DSI indicating an estimated measure of impact to the networked computing environment which would be caused by performance of the example action after enabling the example action. Output of an interactive user interface including information representing the action and the estimated DSI is caused.

Abstract: Disclosed are systems and methods that generate individual semantic layers for different object types within an environment. For each object type semantic layer and for an aerial vehicle, a semantic configuration space may be formed that indicates all valid aerial vehicle configurations in which the aerial vehicle does not collide with objects of the object type that exists in the environment. Finally, a combined configuration space may be formed by overlaying multiple semantic configuration spaces for object types known to be within the environment. The resulting combined configuration space indicates a common area in which the aerial vehicle can navigate according to the configuration of the common area and not collide with any objects of the object types represented by the combined semantic configuration spaces.

Abstract: Vibrometric signatures of portions of a vehicle, such as a joint between two or more aspects of the vehicle, may be generated and used to make determinations regarding the integrity or suitability of the vehicle for one or more missions. Vibrometric signatures are generated based on imaging data captured with the vehicle subjected to excitation over a range of frequencies, and power levels of vibrations of the vehicle are calculated based on the images. Pixels corresponding to surfaces on either side of a joint are selected and vibrometric signatures calculated for such surfaces are compared to determine whether the joint adequately transfers vibrations, or whether the joint is sufficiently tight. Where the vibrometric signatures are sufficiently similar to one another, the joint may be presumed to be tight. Where the vibrometric signatures are not sufficiently similar to one another, the joint is presumed to require maintenance or inspection.

Abstract: Systems, methods, and computer-readable storage media for web-scale visual search capable of using a combination of visual input modalities are provided. An edgel index is created that includes shape-descriptors, including edgel-based representations, that correspond to each of a plurality of images. Each edgel-based representation includes pixels that depicts edges or boundary contours of an image and is created, at least in part, by segmenting the image into a plurality of image segments and performing a multi-phase contour detection on each segment.

Abstract: Systems, methods, and computer-readable storage media for web-scale visual search capable of using a combination of visual input modalities are provided. An edgel index is created that includes shape-descriptors, including edgel-based representations, that correspond to each of a plurality of images. Each edgel-based representation includes pixels that depicts edges or boundary contours of an image and is created, at least in part, by segmenting the image into a plurality of image segments and performing a multi-phase contour detection on each segment.

Abstract: Systems, methods, and computer-readable storage media for web-scale visual search capable of using a combination of visual input modalities are provided. An edgel index is created that includes shape-descriptors, including edgel-based representations, that correspond to each of a plurality of images. Each edgel-based representation includes pixels that depicts edges or boundary contours of an image and is created, at least in part, by segmenting the image into a plurality of image segments and performing a multi-phase contour detection on each segment.

Abstract: Three-dimensional model validation. As a part of the model validation, one or more three-dimensional models of a structure and at least one real world image of the structure is accessed. Features of the one or more three-dimensional models of the structure are combined with features of the at least one real world image of the structure to create a composite model of the structure. The composite model is projected into an imaging plane. The composite model of the structure is compared with at least one of the at least one real world images. Based on the results of the comparing, the accuracy of the one or more three-dimensional models of the structure is determined.

Abstract: Computer-readable media, systems, and methods for flexible matching with combinational similarity are described. In embodiments, an object image is received, a query image is received, and the query image is compared with the object image. In various embodiments matching information is determined based upon combinational similarity and the matching information is presented to a user. In various embodiments, comparing the query image with the object image includes dividing the object image into agents, creating a gradient histogram for the agents, determining map areas for the query image, creating a gradient histogram for the map areas, and creating a similarity array for each of the agents. Further, in various embodiments, determining matching information includes creating a combinational array by combining the similarity arrays for each agent and determining whether the combinational array includes a peak value.

Abstract: Viewing apparel in a store or a catalog may not show a purchaser how the item will look in different light or settings. A user may select elements of a scene, such as a setting, a mannequin, a pose for the mannequin, and apparel/accessories from a web browser-based application. The selected elements are processed by a hierarchy of services that first divide the scene into component elements, render each element, and return the result to a composition server that combines and flattens the renderings into a 2D image. The 2D image is viewable on any platform or browser without the need for special graphics hardware.

Abstract: Computer-readable media, systems, and methods for flexible matching with combinational similarity are described. In embodiments, an object image is received, a query image is received, and the query image is compared with the object image. In various embodiments matching information is determined based upon combinational similarity and the matching information is presented to a user. In various embodiments, comparing the query image with the object image includes dividing the object image into agents, creating a gradient histogram for the agents, determining map areas for the query image, creating a gradient histogram for the map areas, and creating a similarity array for each of the agents. Further, in various embodiments, determining matching information includes creating a combinational array by combining the similarity arrays for each agent and determining whether the combinational array includes a peak value.

Abstract: Computer-readable media, systems, and methods for flexible matching with combinational similarity are described. In embodiments, an object image is received, a query image is received, and the query image is compared with the object image. In various embodiments matching information is determined based upon combinational similarity and the matching information is presented to a user. In various embodiments, comparing the query image with the object image includes dividing the object image into agents, creating a gradient histogram for the agents, determining map areas for the query image, creating a gradient histogram for the map areas, and creating a similarity array for each of the agents. Further, in various embodiments, determining matching information includes creating a combinational array by combining the similarity arrays for each agent and determining whether the combinational array includes a peak value.

Abstract: Three-dimensional model validation. As a part of the model validation, one or more three-dimensional models of a structure and at least one real world image of the structure is accessed. Features of the one or more three-dimensional models of the structure are combined with features of the at least one real world image of the structure to create a composite model of the structure. The composite model is projected into an imaging plane. The composite model of the structure is compared with at least one of the at least one real world images. Based on the results of the comparing, the accuracy of the one or more three-dimensional models of the structure is determined.

Abstract: Computer-readable media, systems, and methods for flexible matching with combinational similarity are described. In embodiments, an object image is received, a query image is received, and the query image is compared with the object image. In various embodiments matching information is determined based upon combinational similarity and the matching information is presented to a user. In various embodiments, comparing the query image with the object image includes dividing the object image into agents, creating a gradient histogram for the agents, determining map areas for the query image, creating a gradient histogram for the map areas, and creating a similarity array for each of the agents. Further, in various embodiments, determining matching information includes creating a combinational array by combining the similarity arrays for each agent and determining whether the combinational array includes a peak value.