Abstract: Techniques for machine learning prediction using edge devices are described. In some examples a method of use includes receiving a request to load a second model onto an edge device while a first model has already been loaded on the edge device, wherein the second model and the first model share an external handle; loading at least one instance the second model into memory of the edge device; and after the second model has been loaded into memory of the edge device, directing a prediction request to the shared external handle to the second model instead of the first model.

Abstract: Techniques for predicting an answer to a question using a machine learning model are described. In some examples, the model predicts one or more answers to the question by: predicting at least two answers to the question using a first component of the question-answer model from a set of passages, generating, using a second component of the question-answer model, at least one question for each of the predicted at least two answers, and performing roundtrip predictions until each generated question only has one answer.

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: A first distributed unit (DU) of a radio-based application determines that traffic associated with a particular user equipment device (UED) is to be handed over to a second DU. The first DU transmits contents of a buffer, including a data block transmitted from the UED, to the second DU using a direct network channel established between the DUs. The second DU stores the contents in a second buffer, and uses the contents to perform a retransmission operation associated with the traffic of the UED.

Abstract: Systems and methods for device integration in augmented reality environments are disclosed. For example, prior to an AR session being initiated, smart devices within a physical environment where the AR session is to be utilized may be identified and selected based on the functionality of the smart devices and the capabilities of an AR application associated with the AR session. Location information about the smart devices may also be determined and this information may be utilized to generate configuration data, which may be used by the AR application to cause device state transitions on the smart devices in response to events that occur in an AR environment of the AR session.

Abstract: A system that provides responses to requests obtains a key that is used to digitally sign the request. The key is derived from information that is shared with a requestor to which the response is sent. The requestor derives, using the shared information, derives a key usable to verify the digital signature of the response, thereby enabling the requestor to operate in accordance with whether the digital signature of the response matches the response.

Abstract: User enrollment to a biometric identification system begins on selected general input devices (GID) such as smartphones. The user may enter identification data (e.g. name) and use a red-green-blue (RGB) camera of the GID to acquire a first image (e.g. hand). The first image is processed using both a first model to determine a first representation and a second model to determine a second representation. Upon presentation of a hand at a biometric input device, a second image is acquired using a first modality and a third image is acquired using a second modality. The second image is processed using the first model to determine a third representation. The third image is processed using the second model to determine a fourth representation. Given a match between both the first and third representations, as well as the second and fourth representations, enrollment is completed by storing the third and fourth representations.

Abstract: Described implementations monitor potential voltage at a location to determine device usage at the location. The implementations utilize a plug-in energy sensor that is plugged directly into any electrical outlet at the location and measures deviation in voltage at the location. Once plugged into an electrical outlet, the plug-in energy sensor monitors one or more of the lead line and ground, the neutral line and ground, and/or the ground line with respect to a reference for changes in potential voltage at the location. Different devices and different device states of those devices generate different signatures of voltage as they operate, thereby making the signatures unique, or almost unique, to different device types, different devices, and/or different device states. By monitoring for signatures in the voltage, operation of devices in different device states at the location can be determined, along with start times and stop times of those states associated with the various devices.

Abstract: A first extension resource group (ERG) of a provider network, comprising a radio-based application pipeline processing server with a network accelerator card, is configured at an external premise. A first set of network functions of a radio-based application is executed at the card. A second ERG is configured at the premise, and a second set of network functions of the radio-based application is executed at the second ERG.

Abstract: A target voice dataset may be augmented using speech prediction. Encoder and decoder models may be trained to encode audio data into encoded speech data, and convert it back to audio. The encoded units may include semantic information (e.g., phonemes and/or words) as well as feature data indicating prosody, timbre, speaker identity, speech style, emotion, etc. of speech. An acoustic/semantic language model (ASLM) may be configured to predict encoded speech data in a manner analogous to a language model predicting words; for example, based on preceding encoded speech data. The models may be used to generate synthesized speech samples having voice characteristics (e.g., feature data) similar to those of the target voice dataset. The augmented dataset may be used to train a text-to-speech (TTS) model to reproduce the target voice characteristics, and may improve performance of the TTS model over training with only the original target voice dataset.

Abstract: Devices and techniques are generally described for object matching in image data. In various examples, first image data and second image data may be received. A first feature map representing the first image data and a second feature map representing the second image data may be generated. The first feature map and second feature map may be combined and a first and second segmentation mask may be generated using the combined feature map. The first segmentation mask may be used to filter the first feature map to generate a filtered representation. The second segmentation mask may be used to filter the second feature map to generate a filtered representation. A determination may be made that a first object depicted in the first image data corresponds to a second object depicted in the second image data using the filtered representations.

Abstract: Systems and methods for providing vehicle software deployment plans that include one or more fallback deployment plans are disclosed. In some embodiments, a vehicle software deployment system determines a deployment plan for deploying one or more software applications one or more electronic control units (ECUs) of a vehicle. Additionally, for one or more failure scenarios, the vehicle software deployment system determines one or more respective fallback deployment plans, such as a deployment plan that assumes one of the ECUs of the vehicle has failed. A deployment plan bundle is provided to the vehicle, comprising a primary deployment plan as well as the one or more fallback deployment plans. In the event that one of the failure scenarios takes place with respect to the vehicle, a deployment agent of the vehicle automatically deploys one of the fallback deployment plans that was provided with the deployment bundle.

Abstract: A system is provided for a self-learning policy engine that can be used by various spoken language understanding (SLU) processing components. The system also provides for sharing contextual information from processing performed by an upstream SLU component to a downstream SLU component to facilitate decision making by the downstream SLU component. The system also provides for a SLU component to select from a variety of actions to take. A SLU component may implement an instance of the self-learning policy that is specifically configured for the particular SLU component.

Abstract: Techniques for understanding classification by a machine learning model are described. A system determines a portion of a user input that causes a machine learning model to classify the user input to a particular classification. The system may generate output data according to the determined portion of the user input. In some cases, the system may generate, using the determined portion, output data explaining processing performed with respect to the user input. In other cases, the system may generate output data requesting feedback. In yet other cases, the system may change a domain, an intent or other classification determined for the user input based on the determined portion.

Abstract: Techniques for altering default language, in system outputs, with language included in system inputs are described. A system may determine a word(s) in user inputs, associated with a particular user identifier, correspond to but are not identical to a word(s) in system outputs. The system may store an association between the user identifier, the word(s) in the user inputs, and the word(s) in the system outputs. Thereafter, when the system is generates a response to a user input, the system may replace the word(s), traditionally in the system outputs, with the word(s) that was present in previous user inputs. Such processing may further be tailored to a natural language intent.

Abstract: A system configured to perform style-aware listener animation. By representing different listening styles (e.g., facial expressions) using an embedding space, a single model can be trained to generate unique facial animations for a number of distinct listeners. Thus, individual listening styles can be associated with a listener identifier, enabling the system to (i) animate a plurality of different listeners with unique nonverbal behavior and/or (ii) select a particular listener identifier or desired type of listener style with which to animate. This enables the model to be generalized to new listeners to generate additional listener facial responses without needing training data for each new listener. The model may process a listener representation style or listener identifier, along with input data corresponding to a speaker talking, to generate unique facial animation responsive to the speech.

Abstract: A response initiation time target based at least in part on a service level agreement is assigned to a task requested from a network-accessible service. A deadline parameter of a thread identified to perform sub-tasks of the task at a service logic implementation node is set based on the target. The thread is scheduled for execution using an operating system thread scheduling algorithm which selects threads for execution based on their deadline parameters. After a first sub-task of thread is completed, a response is sent to the requester of the task.

Abstract: Systems and methods are disclosed for automated self-calibration and maintenance for container shuttles. An example system for automated repairs may include a track, a shuttle coupled to a permanent magnet, the shuttle configured to move along the track, and an electromagnet disposed along the track, where the shuttle is propelled via interaction between the electromagnet and the permanent magnet. The system may include a maintenance area accessible via the track, and a set of robotic manipulators disposed about the maintenance area, the set of robotic manipulators configured to perform automated maintenance actions on the shuttle. The system may include a first controller configured to determine that the shuttle is in need of maintenance, and automatically direct the shuttle to the maintenance area.

Abstract: Techniques for enabling access in a multi-assistant speech processing system are described, where a first assistant system may use components of a second assistant system as data processing components. Runtime operational data and user input data related to the first assistant may be kept separate from the processing data and input data related to the second assistant by propagating a first account ID, for user inputs directed to the first assistant, through the processing pipeline, and using a second account for user inputs directed to the second assistant. A mapping between the first account ID and the second account ID may be accessible to a select number of system components. Handoffs between the two assistants are handled in a manner where data related to one assistant is not accessible by the other assistant.

Abstract: Approaches presented herein relate to visibility into network traffic associated with one or more groups of network resources. Approaches provide for automatic deployment and configuration of one or more visibility pods for individual groups of networked resources. A visibility pod may include at least one virtual compute instance to execute at least one analytical tool or application. A copy of network traffic for the monitored resources can be directed to an endpoint for a respective visibility pod, which can extract information (e.g., specific metadata about the traffic) to be analyzed by the analytical tools or applications. The results of the analysis can be written to a log file and/or data repository, and in some instances may be used to generate security alerts or other such notifications. A visibility pod can be automatically deployed and configured for any group of networked resources from which network traffic can be directed.