Abstract: A machine learning model analyzes text describing an item to determine portions of the text that correspond to multiple characteristics of the item. A first set of embeddings that represent the text describing the item is determined. A second set of embeddings that represent text indicating the characteristics is determined. The second set of embeddings includes a token for each characteristic that is used to indicate tokens that are associated with a particular characteristic. The first set of embeddings and portions of the second set of embeddings for a particular characteristic are used to determine a set of interaction embeddings for each characteristic by determining an element-wise product. These interaction embeddings are analyzed to determine label predictions indicating text that is associated with each characteristic. Text for multiple characteristics may therefore be identified using a single pass rather than multiple passes.

Abstract: Devices and techniques are generally described for nearest device arbitration. In various examples, a first device may receive first audio data representing a wakeword spoken by a first speaker at a first time. In some examples, a second device may receive second audio data representing the wakeword spoken by the first speaker at the first time. In some cases, the first device may generate first feature data representing the first audio data and the second device may generate second feature data representing the second audio data. In various examples, a machine learning model may use the first feature data and the second feature data to generate first prediction data representing a prediction that the first device is closer to the first speaker than the second device.

Abstract: The present disclosure presents a generative model configured to receive input regarding an item in two different modalities, such as text data and non-text data (including, for example, image or audio data), in order to generate output regarding the item that is determined based on a combination of both modalities' input. Specific relative positional and token type embeddings may be employed in an encoder portion of an encoder-decoder arrangement. An associated decoder may be trained to generate new text corresponding to diverse tasks based on the encoded representation of the two inputs as generated within the encoder. For example, the decoder may be utilized to generate attributes regarding the input item, auto-complete or auto-correct a title or description of the item, among other uses.

Abstract: Systems and methods are described relating to aggregating log anomalies. In some examples, a plurality of log anomaly instances may be obtained, from a log anomaly detector, where individual instances are associated with a first log anomaly type and a first anomalous log event. Log anomaly instances associated with the first log anomaly type and the first anomalous log event may be combined into a first log anomaly class. The first log anomaly class may be combined with a second log anomaly class, including log anomaly instances associated with the first anomalous log event and a second log anomaly type, into a log anomaly group, which may correlate the occurrences of the first and second anomaly types to the same first anomalous log event over a period of time. An indication of the log anomaly group may then be output.

Abstract: Systems and methods in provide approaches for channel fitting to determine a particular number of resources, such as cores, that may be used for one or more processing jobs for a particular input channel. A given channel may be evaluated to determine performance affecting parameters and those parameters may be evaluated against previously known hardware configurations or against a historical heuristic dataset. A configuration for the channel may be selected where particular cores of a set of resources are assigned to the channel to perform one or more operations, such as transcoding operations. As a result, empirically determined or historical data may be used to efficiently allocate resources for different transcoding operations to provide both predictable performance and high utilization.

Abstract: Automatic partitioning of a machine learning model may be performed for training across multiple processing devices. A training job for a machine learning model may specify a number of partitions for a machine learning model. An optimization parameter may be determined for the machine learning model. Different partitions of the machine learning model to be trained across multiple processing devices may be determined based on the specified number of partitions and the optimization parameter. A schedule for executing the training job may be generated according to the respective partitions of the machine learning model. The training job may be executed according to the schedule.

Abstract: Systems and methods are disclosed for harvesting electrical energy from mechanical components of hard disk drives (HDDs) in a data storage system and propagating the electrical energy to devices outside of the HDDs. A power distribution board (PDB) may be coupled to a plurality of HDDs and used to detect a voltage drop on a connection between the PDB and the HDDs indicative of a power loss condition, and, in response, enable the flow of electrical energy from the HDDs to the PDB. The electrical energy from the HDDs may be converted for use by the PDB and/or distribution to other components of the data storage system.

Abstract: Systems and methods are provided to improve system performance when multiple transaction retry events associated with transaction requests from requester nodes to completer nodes are detected. A retry monitor can monitor the transaction retry events associated with the transaction requests to provide retry information. An intervention level generator can receive information about the transaction retry events and determine an intervention level from a plurality of intervention levels based on the retry information and a retry configuration. Each requester node can be coupled to a regulator to regulate the transactions being requested by that requester node based on the intervention level and a regulator configuration, which can allow the corresponding completer nodes to complete the outstanding transactions and reduce the occurrence of retry events.

Abstract: A video packaging and origination service can process requests for content segments from requesting user devices. The video packaging and origination service can processing video attributes, audio attributes and social media feeds to dynamically determine insertion points for supplemental content. Additionally, the video packaging and origination service can identify supplemental content utilizing the same attribute information.

Abstract: Database query processing may be scaled using additional processing clusters. A database query is received at a processing cluster. A determination is made as to whether additional processing clusters will be used to process the database query. Operations to cause compute nodes of the processing cluster to instruct operations at the additional processing clusters are included in a plan generated to perform database queries determined to use additional processing clusters. The plan is executed to be perform the database query causing compute nodes of the processing cluster to send instructions to corresponding additional processing clusters in order to generate and return a response to the database query.

Abstract: A system configured to perform audio processing with adaptive multi-stage output gains. For example, an Audio Front End (AFE) component may generate a first output using a fixed gain value in order to improve device arbitration and a second output using an adaptive gain value in order to improve wakeword detection. A wakeword engine may process the second output to determine that a wakeword is present along with start/end times of the wakeword. The AFE component can use the start/end times to determine an amount of wakeword energy represented in the first output, which is sent to a remote device for device arbitration. The AFE component can also use the start/end times to determine an amount of wakeword energy represented in the second output, which can be used to determine the adaptive gain value that is unique to the device.

Abstract: Systems and methods are disclosed to update ads on user devices that may connect to a network only intermittently by predicting and tracking various metrics associated with ads delivered to the user devices. An example method may include determining a first set of ads that were presented at a first semi-connected device while the first semi-connected device was disconnected from a network, determining a first device identifier associated with the first semi-connected device, and determining, based at least in part on the first device identifier, a first estimated number of ad impressions for a first ad. Example methods may include determining, based at least in part on the first estimated number of ad impressions, that the first semi-connected device is to be updated with the first ad, causing the first ad to be sent to the first semi-connected device, and causing at least one ad of the first set of ads to be removed from the first semi-connected device.

Abstract: Systems and methods are disclosed for seamless disengagement of shuttles powered by electromagnets. An example system may include a frame coupled to a track, a first conveyor disposed along the first side of the track, a second conveyor disposed along the second side of the track, and an actuator configured to drive the first conveyor and the second conveyor, where the first conveyor and the second conveyor are configured to engage a first bar disposed on a lower surface of the shuttle. The system may be configured to (i) move the shuttle from a first location along the track to a second location along the track, and (ii) pause movement of the shuttle between the first location and the second location.

Abstract: Systems and methods provide for submission, verification, and validation of one or more code packages associated with provisioning logic for third-party applications. A user may request verification as a publisher and submit a code package for use with a third-party application within a resource provider environment. The code package may be tested and, upon passing, be published to a public repository for discovery and use by other users. The code package may form part of an extension that is integrated into a template that enables automated provisioning of resources and applications.

Abstract: This disclosure describes a system for associating multiple totes with a single profile so that items placed into any of the multiple totes are identified on a profile item identifier list. For example, if two users located in a materials handling facility are picking items that are to be consolidated or otherwise treated together, each user may select a different tote and move separately through the materials handling facility. As either user picks items and places those items into a corresponding tote, a profile item identifier list is updated to include an item identifier for each picked item.

Abstract: Described herein are approaches for generating a new queue based on an existing queue. This may include receiving a request to transfer the existing queue from a first device to a second device. A set of move criteria may be evaluated using a playback context, a user profile, a configuration associated with the second device, and/or a level of access constraints. Depending on the results of the evaluation, the existing queue may be completely reformulated to define the new queue. The second device may then be instructed to play the new queue.

Abstract: A wearable audio output device (e.g., headphones) having an open design that allows ambient noise to pass to a listener without physically isolating the listener from a surrounding environment. The device may include an open earcup design that may partially or completely surround the listener's ear, and in some examples a portion of the listener's head may be uncovered by the open earcup. To improve comfort, the device includes a floating audio component configured to generate output audio in a direction of the listener's ear without contacting the listener's ear. To reduce an amount of ambient noise, the device may be configured to perform active noise cancellation (ANC) processing using feedforward and/or feedback microphones. The device may include an acoustic structure configured to direct the output audio in the direction of the listener's ear and/or position the feedback microphone(s) closer to the listener's ear.

Abstract: Integrated cooling structures and methods may comprise a motor, a propeller, and a component having at least one structural feature configured to receive or channel airflow to generate a positive pressure region at a first end of the motor. During operation of the motor and propeller, a negative pressure region may be generated at a second end of the motor proximate the propeller. The positive and negative pressure regions may generate a pressure differential or gradient between ends of the motor, such that a cooling airflow may be generated that flows through an interior of the motor from the positive pressure region to the negative pressure region. The cooling airflow may increase cooling of internal components of the motor, and consequently improve various operational capabilities, performance, and reliability of the motor and associated vehicle, machine, or system.

Abstract: Session-specific edit recommendations may be made for editing a code file. After a code editing session is started, code file edits may be captured. A machine learning technique may be applied to learn the code edits and recommend alternative code portions for portions of the code file during the code editing session. The recommendations may be provided and accepted, or not, via an interface of a code editor application.

Abstract: This disclosure is directed to, in part, a processing pipeline for detecting predefined activity using image data, identifying an item-of-interest and a location of the item-of-interest across frames of the image data, determining a trajectory of the item-of-interest, and determining an identifier of the item-of-interest and an action taken with respect to the item-of-interest. The processing pipeline may utilize one or more trained classifiers and, in some instances, additional data to identify items that are placed into or removed from a tote (e.g., basket, cart, or other receptacle) by users in material handling facilities as the users move around the material handling facilities.