Abstract: Computing units provided at local sites or edge locations are programmed to execute conversational tools that generate pertinent, domain-specific responses to queries received from workers at such sites or locations. The conversational tools are large language models that are trained with domain-specific knowledge documents. Data representing queries are received from workers at such sites or locations and provided as inputs to the conversational tools along with text representing nearest knowledge documents from a knowledge base associated with the domain, as well as contextual data. Responses identified based on outputs received from the conversational tools in response to the inputs are provided to the workers that generated the queries.

Abstract: A process can include obtaining a plurality of data points each associated with a respective edge device of a fleet of edge devices, each respective edge device associated with an edge site location or edge device asset group. The plurality of data points are stored to a fleet map data catalog and a filtering selection for viewing a filtered subset of the fleet map data catalog is received, indicating a selected geographic view area and selected edge device types from a plurality of edge device types. Data points corresponding to the filtered subset are obtained from the fleet map data catalog using the filtering selection. A fleet map GUI view is generated using the data points corresponding to the filtered subset, the fleet map GUI view comprising a converged geographic map of the selected geographic view area, with data points are rendered at corresponding locations within the converged geographic map.

Abstract: An apparatus can be configured to receive monitoring information associated with a machine learning (ML) or artificial intelligence (AI) workload implemented by an edge compute unit of a plurality of edge compute units. Status information corresponding to a plurality of connected edge assets can be received, the plurality of edge compute units and connected edge assets included in a fleet of edge devices. A remote fleet management graphical user interface (GUI) can display a portion of the monitoring or status information for a subset of the fleet of edge devices, based on a user selection input, and can receive a user configuration input indicative of an updated configuration for at least one workload corresponding to a pre-trained ML or AI model deployed on the at least one edge compute unit. A cloud computing environment can transmit control information corresponding to the updated configuration to the at least one edge compute unit.

Abstract: Systems and techniques are provided for processing streaming data. For example, a process can include obtaining, by an edge device, a plurality of data streams corresponding to objects, activities, or events registered in an edge environment associated with the edge device. One or more machine learning networks implemented on the edge device can generate a plurality of features and/or embeddings corresponding to each respective data stream of the plurality of data streams. The edge device can determine a subset of salient content from the plurality of data streams, based on analyzing the plurality of features and/or embeddings generated for each respective data stream. The edge device can generate index information corresponding to the determined subset of salient content. The subset of salient content and the generated index information can be stored locally at the edge device for search and retrieval based on local queries received at the edge device.

Abstract: A process can include receiving monitoring information associated with a machine learning (ML) or artificial intelligence (AI) workload implemented by an edge compute unit of a plurality of edge compute units. Status information corresponding to a plurality of connected edge assets can be received, the plurality of edge compute units and connected edge assets included in a fleet of edge devices. A remote fleet management graphical user interface (GUI) can display a portion of the monitoring or status information for a subset of the fleet of edge devices, based on a user selection input, and can receive a user configuration input indicative of an updated configuration for at least one workload corresponding to a pre-trained ML or AI model deployed on the at least one edge compute unit. A cloud computing environment can transmit control information corresponding to the updated configuration to the at least one edge compute unit.

Abstract: A reliability map is generated by superimposing or associating populations or other intrinsic data is superimposed or associated with a geographic map of a region, which is divided into a grid having cells of uniform size. Densities of towns, cities or other geospatial areas are determined and assigned to cells of the grid, which have sizes corresponding to minimum dimensions of a corridor required for travel by an aerial vehicle. When a mission requiring travel from an origin to a destination within the region is identified, one or more paths of a safe route between the origin and the destination are selected based on the reliability map. The safe route is selected to avoid areas of high population density or locations of critical infrastructure such as schools, hospitals or public safety buildings.

Abstract: Bid values submitted for various keywords can take into account the recovery propensity between paid search and organic search. When submitting a bid to a search engine provider for a keyword, an entity may get a certain level of performance in return. If not submitting a bid, however, the entity will likely still get some level of performance, although likely less than for paid search. In order to optimize for a parameter such as impressions, purchases, or profit, the recovery propensity can be taken into account in order to adjust the bid price, taking into account the relative performance of paid and organic search and then optimizing for the determined goal. Organic search data in some embodiments can be obtained through testing or modeling, or a combination thereof.

Abstract: Passive brushes formed from plastics, rubbers or other suitable materials may be inserted into openings of aerial vehicles or other machines and subjected to excitation at natural frequencies of vibration of the passive brushes. Where surfaces of the passive brushes are in contact with surfaces of components that are fouled by dust, dirt, grime, or other substances, the vibration of the passive brushes resulting from the excitation causes the dust, dirt, grime or other substances to be released from such surfaces. The passive brushes may have dimensions or shapes that are selected to correspond to internal dimensions of the aerial vehicles or specific components, and may be formed by 3D printing, nanolithography, or in any other suitable manner.

Abstract: A customer who visits an online marketplace to view information regarding an item formed from fabric may request and receive instructions for producing a sample of the fabric that mimics the sensations of touch and feel of the fabric. The sample may be produced using an automated fabricator, such as a 3D printer, associated with a client device from which the customer made the request. The instructions may include or describe a digital model of the fabric, customized for the automated fabricator. In particular, the sample may be formed from various plastics or other materials that are accessible to the automated fabricator, and need not be formed from the actual materials that are included in the fabric.

Abstract: Techniques for managing notifications may be described. In an example, the notifications may relate to an item and may be provided to a user device. A passive device may be associated with the item. A delivery management device at a particular location may detect the proximity of the passive device. The delivery management device may cause a notification to be sent to the user device based on the proximity. The notification may describe that the item may have been detected at the particular location.

Abstract: An occupancy map of a region is generated for a vehicle by dividing the region into cells and calculating occupancy metrics for each of the cells based on building footprints or other data. Cells having high occupancy metrics are labeled as obstructed, and cells having low occupancy metrics are labeled as free of obstructions. Cells that may not be labeled as obstructed or free based on their occupancy metrics are subdivided, and occupancy metrics are calculated for cells formed from the subdivision. Occupancy metrics may be calculated for cells, and the cells may be subdivided, for as long as the cells formed from the subdivision have areas greater than an area defined based on operational capabilities of the vehicle. The occupancy map is used to determine routes through the region for the vehicle.

Abstract: Systems and methods related to designation, design, and service of replaceable components of aerial vehicles may include designating a component as a replaceable component based on a determined service frequency, designing the replaceable component for incorporation into an aerial vehicle based on the service frequency, and servicing the replaceable component according to the service frequency. In this manner, a replaceable component having a relatively high service frequency may be incorporated into an aerial vehicle at a relatively more accessible location, and a replaceable component having a relatively low service frequency may be incorporated into an aerial vehicle at a relatively less accessible location, thereby facilitating efficient and reliable maintenance of replaceable components.

Abstract: A pervasive search architecture that indexes personal content of a querying user and made accessible to the user by other users. A compute node of a personal content location facilitates index generation and serve of the index. The index is generated for personal content stored at the personal content location. For a given content location, the index may encapsulate content stored in a set of locations with access permissions. The indexing application runs periodically at the personal content location and incrementally indexes content that is added to the shared locations. The same application allows the user to configure locations with the desired access permissions for participation in the search.

Abstract: Systems and methods relating to an end-of-arm-tool that can be used in connection with the automated handling of vehicles, such as unmanned aerial vehicles (UAV), are disclosed. The described systems and methods can include an end-of-arm-tool which may include a load cell coupled to an end effector, such that forces and torques exerted on the end effector are translated onto the load cell. The measurement of forces and torques exerted on the end effector can facilitate determining various information in connection with the aerial vehicle, such as inertial properties or parameters associated with the aerial vehicle, the quality of the engagement between the end effector and the aerial vehicle, as well as diagnostic information in connection with the aerial vehicle. Additionally, the use of a load cell to measure forces and torques exerted on the end effector can eliminate the need to utilize traditional contact sensors typically required on the contact surfaces of an end-of-arm tool.

Abstract: A population density map of a region is generated by dividing the region into cells and allocating a population of the region only to the cells that are accessible to people, or are believed to be populated. Each of the cells is classified based on one or more ground features of the cells, and an adjustment factor for each of the cells is determined based at least in part on the classifications. Equal shares of the population of the region are allocated to each of the cells that is accessible or populated, and the equal shares are multiplied by the adjustment factors determined for the respective ones of the cells to calculate a population for each of such cells.

Abstract: Systems and techniques are disclosed for a blockchain network based on machine learning-based proof of work. One of the methods includes accessing a blockchain associated with a blockchain network, and obtaining a first error value specified in a block of the blockchain, the first error value being associated with a machine learning model identified in the block, and the blockchain recording machine learning models. A new machine learning model associated with a second error value is determined, with the second error value being less than the first error value. A block proposal identifying the new machine learning model is generated, the block proposal specifying the first error value. Transmission of the block proposal to other entities is caused. In response to greater than a threshold percentage of the entities approving the block proposal, inclusion of the block proposal in the blockchain is caused.

Abstract: Deep feature extraction and training tools and processes may facilitate extraction and understanding of deep features utilized by deep learning models. For example, imaging data may be tessellated and masked to generate a plurality of masked images. The masked images may be processed by a deep learning model to generate a plurality of masked outputs. The masked outputs may be aggregated for each cell of the tessellated image and compared to an original output for the imaging data from the deep learning model. Individual cells and associated image regions having masked outputs that correspond to the original output may comprise deep features utilized by the deep learning model.

Abstract: Techniques are generally described for causal impact estimation using machine learning. A first machine learning model is trained using non-treatment variables during training. A second machine learning model uses learned weights from the first machine learning model for non-treatment variables and is trained on one or more treatment variables. The second machine learning model estimates outcomes based on the presence or absence of an event represented by the treatment variable. Selection bias is reduced by warm-starting the second machine learning model with non-treatment variable weights learned during training of the first machine learning model.

Abstract: Aspects of the present disclosure relate to machine learning techniques for identifying the incremental impact of different past events on the likelihood that a target outcome will occur. The technology can use a recurrent neural network to analyze two different representations of an event sequence—one in which some particular event occurs, and another in which that particular event does not occur. The incremental impact of that particular event can be determined based on the calculated difference between the probabilities of the target outcome occurring after these two sequences.

Abstract: Aspects of the present disclosure relate to machine learning techniques for continuous implementation and training of a machine learning system for identifying the natural language meaning of visual content. A computer vision model or other suitable machine learning model can predict whether a given descriptor is associated with the visual content. A set of such models can be used to determine whether particular ones of a set of descriptors are associated with the visual content, with the determined descriptors representing a meaning of the visual content. This meaning can be refined based on a multi-armed bandit tracking and analyzing interactions between the visual content and users associated with certain personas related to the determined descriptors.