Abstract: Techniques for performing outputting additional content associated with but nonresponsive to an input command are described. A system receives input data from a device. The system determines an intent representing the input data and receives first output data responsive to the input data. The system determines, based on context data, that additional content associated with the first output data but nonresponsive to the input data should be output. The system receives second output data associated with but nonresponsive to the input data thereafter. The system then presents first content corresponding to the first output data and second content corresponding to the second output data.

Abstract: An autonomous mobile device (AMD) interacts with a user to provide tasks such as conveniently displaying information on a screen and moving with the user as they move. The AMD determines an area, or bounding box, of a user appearing within images obtained by a camera that is mounted on the AMD. A preferred area, with respect to the images, such as a center of the image, is specified to provide desired framing of images. As images are acquired by the camera, a difference between the bounding box and the preferred area is determined. Based at least in part on this difference, instructions are determined to move one or more of the cameras or the entire AMD to try and reframe the bounding box in subsequent images closer to the preferred area. Other factors, such as the user being backlit, may also be considered in determining the instructions.

Abstract: A generative adversarial network (GAN) may be implemented to recognize named entity types in detection of sensitive information in datasets. The GAN may include a generator and a discriminator. The generator may be trained to produce synthetic data to include information that simulates named entity types representing the sensitive information. The discriminator may be fed with real data that are known to include the sensitive information (as positive examples), together with the synthetic data that simulate the sensitive information (as negative examples), to train to classify the real vs. synthetic data. In field operations, the discriminator may be deployed to perform named entity type recognition to identify data having the sensitive information. The generator may be deployed to provide anonymous data in lieu of real data to facilitate sensitive information sharing and disclosure.

Abstract: This disclosure relates to answer prediction in a speech processing system. The system may disambiguate entities spoken or implied in a request to initiate an action with respect to a target user. To initiate the action, the system may determine one or more parameters; for example, the target (e.g., a contact/recipient), a source (e.g., a caller/requesting user), and a network (voice over internet protocol (VOIP), cellular, video chat, etc.). Due to the privacy implications of initiating actions involving data transfers between parties, the system may apply a high threshold for a confidence associated with each parameter. Rather than ask multiple follow-up questions, which may frustrate the requesting user, the system may attempt to disambiguate or determine a parameter, and skip a question regarding the parameter if it can predict an answer with high confidence. The system can improve the customer experience while maintaining security for actions involving, for example, communications.

Abstract: Configurable core domains of a speech processing system are described. A core domain output data format for a given command is originally configured with default content portions. When a user indicates additional content should be output for the command, the speech processing system creates a new output data format for the core domain. The new output data format is user specific and includes both default content portions as well as user preferred content portions.

Abstract: A quantum programming environment may include an assisted composition system to assist the composition of quantum objects. The assisted composition system may receive a partial portion of a quantum object that is being composed but not yet fully composed by a user. The assisted composition system may determine a first abstract representation of the partial portion of the quantum object being composed. The assisted composition system may determine that the first abstract representation resembles at least a first portion of a second abstract representation of a stored quantum object stored in a library for the quantum programming environment. The assisted composition system may obtain a second portion of the stored quantum object from the library and provide it to the user as a next portion to the partial portion of the quantum object being composed.

Abstract: Disclosed herein are techniques for performing multi-layer neural network processing for multiple contexts. In one embodiment, a computing engine is set in a first configuration to implement a second layer of a neural network and to process first data related to a first context to generate first context second layer output. The computing engine can be switched from the first configuration to a second configuration to implement a first layer of the neural network. The computing engine can be used to process second data related to a second context to generate second context first layer output. The computing engine can be set to a third configuration to implement a third layer of the neural network to process the first context second layer output and the second context first layer output to generate a first processing result of the first context and a second processing result of the second context.

Abstract: Two or more personal area networks (or PANs) operating channel hopping schedules may be operated within a vicinity of one another with channel hopping schedules according to IEEE 802.15.4 that are separated by a frequency offset, and collisions between data transmitted by the respective devices of the networks may be avoided. Nodes of a first PAN transmit beacons identifying information regarding a first communications schedule. A node of a second PAN receives the beacon and establishes a second channel hopping schedule at the offset from the first communications schedule. Nodes of PANs may be programmed with communication modes that vary uplink or downlink slots based on quality-of-service or traffic requirements.

Abstract: Systems and methods are provided for management of network segments that cross geographic regions and/or other types of network divisions in a cloud-based network environment. Gateway may manage traffic across regions using routing metadata that includes a segment identifier. The gateways may also signal their routes across regions based on segment data, and implement the signaled routes using segment-based routing policies. Route selection may be performed using optimization data.

Abstract: Techniques to partition a neural network model for serial execution on multiple processing integrated circuit devices are described. An initial partitioning of the model into multiple partitions each corresponding to a processing integrated circuit device is performed. For each partition, an execution latency is calculated by aggregating compute clock cycles to perform computations in the partition, and weight loading clock cycles determined based on a number of weights used in the partition. The amount of data being outputted from the partition is also determined. The partitions can be adjusted by moving computations from a source partition to a target partition to change execution latencies of the partitions and the amount of data being transferred between partitions.

Abstract: In response to determining that a particular sequence of natural language input has been generated by a first entity participating in a multi-interaction dialog, a first representation of accumulated dialog state associated with the sequence is obtained from a machine learning model at an artificial intelligence service. Based on the first representation, a state response entry is selected from a collection of state response entries. The state response entry indicates a mapping between a second representation of accumulated dialog state, and a response recorded in a training example of the model. The recorded response is implemented.

Abstract: A new type system may be added to a type registry for a data processing service. A request to add the new type system may be received that describes the new type system for a data store. The new type system may be used to perform a data processing job that accesses the data store to obtain or store data as a source or target data store.

Abstract: Data transformation workflows may be generated to transform data objects. A source data schema for a data object and a target data format or target data schema for a data object may be identified. A comparison of the source data schema and the target data format or schema may be made to determine what transformations can be performed to transform the data object into the target data format or schema. Code to execute the transformation operations may then be generated. The code may be stored for subsequent modification or execution.

Abstract: A hierarchical embedding model is used to obtain respective language-agnostic embeddings of entity records of a cross-language data set. A plurality of record representation pairs is prepared based at least in part on the language-agnostic embeddings. A machine learning model is trained using the record representations pairs to generate similarity scores for pairs of entity records whose text attributes are expressed in different languages.

Abstract: An apparatus includes a platform or other accessory that is supported by one or more transducer assemblies using piezoelectric transducers. In one implementation, a saddle structure provides mechanical support to a platform while transferring force from a load onto a portion to which a pair of piezoelectric transducers are affixed. As a load on the saddle changes, the resulting change in strain on the portion produces signals from the piezoelectric transducers. In another implementation a beam provides mechanical support to a platform or other accessory. A pair of piezoelectric transducers are affixed near a strain concentrator, such as a hole in the beam, produce signals indicative of a change in load on the beam.

Abstract: A network-accessible machine learning service is provided herein. For example, the network-accessible machine learning service provider can operate one or more physical computing devices accessible to user devices via a network. These physical computing device(s) can host virtual machine instances that are configured to train machine learning models using training data referenced by a user device. These physical computing device(s) can further host virtual machine instances that are configured to execute trained machine learning models in response to user-provided inputs, generating outputs that are stored and/or transmitted to user devices via the network.

Abstract: The present disclosure generally relates to systems and methods for the automatic creation of trace data. A trace generation service generates a trace identifier that will be utilized by a plurality of network services. Individual network services generate log data corresponding to execution of the network-based service. Individual log event in the log data includes information associated with a trace identifier generated by a network-based service during execution of the service. The trace generation service processes the log data based on trace identifiers to identify individual trace span associated with the generated unique trace identifier. The trace generation service constructs trace data associated with the generated unique trace identifier based on the identified trace span data.

Abstract: Techniques for hotspot detection in a dataset are described. A hotspot being a region (or a collection of points) where the value of a function of given any region in the space measures the concentration of points in that region is significantly higher than its other regions of the dataspace. As such, a region that has a denser concentration of points than other regions of the dataspace may be considered a hotspot. In some implementations, hotspot detection includes finding two or more regions to evaluate for high-density in the dataset, a high-density region indicating a potential hotspot and extending a size of the manipulated found two or more regions to determine borders for these regions.

Abstract: Techniques for optimizing and deploying convolutional neural network (CNN) machine learning models for inference using integrated graphics processing units are described. A model compilation system optimizes CNN models using optimized vision-specific operators as well as both graph-level tuning and tensor-level tuning to explore the optimization space for achieving heightened performance. The model compilation system may also implement a heuristic-based two-stage technique for falling back certain operators of CNN models to use CPUs when needed or otherwise beneficial.

Abstract: Generally described, the present application relates to providing a container orchestration service (“service”) that can automatically detect and terminate a failed deployment and automatically roll back to a previous healthy deployment. For example, the service may receive a request to deploy a new user code onto one or more compute instances usable to execute the new user code, where the new user code is an updated version of an old user code to be replaced by the new user code. The service may then attempt to execute a plurality of instances of the new user code on the one or more compute instances, determine, prior to replacing all instances of the old user code to be replaced by the plurality of instances of the new user code, that the deployment of the new user code has satisfied a failed deployment condition, and terminate the deployment of the new user code.

Abstract: Systems and methods for fall detection on uneven surfaces utilizing radar are disclosed. For example, an electronic device may receive reflected energy from surfaces in an environment. Corresponding sensor data associated with an object moving on an uneven surface may be utilized to determine a velocity of the object and a height change of the object. When the velocity indicates that the object has stopped moving toward or away from the sensing device, and the height values indicate a sudden decrease in height, a fall event may be detected.

Abstract: A first analysis based on at least a subset of collected health metrics of a monitored resource is performed at a health monitoring service. If the first analysis indicates that the probability that the monitored resource is in an unhealthy state is above a threshold, a mitigation action that includes directing at least some of the workload away from the monitored resource is implemented. In addition, a remediation analysis of the monitored resource is started to determine, using at least one additional health metric that was not examined in the first analysis, whether a longer-term remediation action is to be initiated.

Abstract: Inferences may be obtained to handle access requests at a non-relational database system. An access request may be received at a non-relational database system. The non-relational database system may determine that the access request uses a machine learning model to complete the access request. The non-relational database system may cause an inference to be generated using data items for the access request as input to the machine learning model. The access request may be completed using the generated inference.

Abstract: Systems and methods are disclosed to provide an autonomous management of communication links between dice on a multi-die assembly. Each die can include a detection unit and a controller to detect a failing communication link and perform link maintenance by directing the communication traffic on the failing link to an operational link before the link fails. Once the failing link has been repaired, the controller can re-direct the traffic back to the repaired link. The controllers on each die can negotiate through a handshake process to provide the continuous operation by switching the communication traffic from the failing link to the operational link, and then from the operational link to the repaired link.

Abstract: Manifest files for proxies within an application mesh or service mesh are created using parallel computing. Virtual services in a mesh definition are identified. For each of the virtual services, one or more virtual nodes that receive traffic from the virtual service are identified. For each of the virtual nodes that receive traffic from the virtual service, an edge is added to a first graph. Virtual nodes that send network traffic to at least one of the virtual services are then identified. For each virtual node that sends traffic to the virtual service, an edge is added to a second graph. Then an optimized per node network topology is generated based at least in part on the first graph and the second graph.

Abstract: Various apparatus are disclosed for lifting heat sinks during insertion of optical modules into transceiver cages (e.g., housings). The apparatus also provide for lifting of the heat sink during removal of the optical modules from the transceiver cages. While the heat sink is lifted during insertion or removal, the heat sink is in contact with the surface of the optical module when the optical module is fully inserted in the transceiver cage. Lifting of the heat sink during insertion and removal of the optical module allows a thermal interface material (TIM) to be placed on the surface of the heat sink in contact with the optical module to improve thermal conduction between the optical module and the heat sink.

Abstract: A multi-modal interactive apparatus provides for output and input using a plurality of modalities. A display presents image output and speakers provide audio output. A sensor module includes sensors such as a microphone array, camera, radar, ambient light sensor, and so forth. Based on how the apparatus is mounted to a supporting structure, in one implementation the overall orientation of the apparatus may be arranged to provide a desired field-of-view (FOV) to the sensors. For example, based on mounting height, the orientation places the sensor module on a top or bottom of the mounted device. Within the sensor module, one or more sensors may be mounted with a pitch angle that directs the FOV to encompass users during typical operation. In another implementation, the sensor module may be repositioned with respect to the apparatus to provide the desired FOV, such as moved from top to bottom.

Abstract: A system that implements a scaleable data storage service may maintain tables in a non-relational data store on behalf of clients. Each table may include multiple items. Each item may include one or more attributes, each containing a name-value pair. Attribute values may be scalars or sets of numbers or strings. The system may provide an API usable to request that values of one or more of an item's attributes be updated. An update request may be conditional on expected values of one or more item attributes (e.g., the same or different item attributes). In response to a request to update the values of one or more item attributes, the previous values and/or updated values may be optionally returned for the updated item attributes or for all attributes of an item targeted by an update request. Items stored in tables may be indexed using a simple or composite primary key.

Abstract: Disclosed techniques provide a permission framework to control access to operations performed by cryptoprocessor. The techniques can identify a permission policy linked to a cryptographic operation. The permission policy can include data identifying the cryptographic operation and data identifying permission information for the cryptographic operation. The permission policy can be evaluated to determine whether to allow or deny execution of the cryptographic operation.

Abstract: Techniques for prioritizing resources of various users, associated with a device, when responding to a user input received from the device are described. When a user input is received from a device, a system may generate a resource list for a group profile (e.g., a household profile) and each user profile (including any guest user profile) associated with the device. Each resource list may include the catalogs of resources (e.g., songs of a playlist, contacts of a contact list, etc.) of the group profile or user profile. The system may also generate a weight matrix including a respective weight for each catalog of each resource list. Various processing components (e.g., an automatic speech recognition component, a natural language understanding component, and an entity resolution component) may process using the resource lists and the weight matrix to determine an output responsive to the user input.

Abstract: An autonomous mobile device (AMD) operating in a first mode moves within a physical space to perform various tasks such as displaying information on a screen, following a user, and so forth. The first mode may involve the AMD moving or maintaining a particular pose. While the AMD is in the first mode, a user may apply an external force to the AMD to reposition the AMD to a desired pose. Application of this external force on the AMD is detected and results in the AMD transitioning to a second mode in which the AMD may be repositioned. While in the second mode, the user may reposition the AMD. The second mode may constrain the magnitude of the resulting movement, preventing the user from moving the AMD too quickly which could damage components within the AMD. Once the external force ceases, the AMD may transition back to the first mode.

Abstract: A system may include first and second speech-processing systems. The first speech-processing system may process received audio data and determine that a command represented therein is associated with a second speech-processing system. The first speech-processing system may send command data to the second speech-processing system and receive response data in return. The first speech-processing system may then process the response data to determine second response data that includes an indication of the second speech-processing system and cause output of audio corresponding to the second response data.

Abstract: Described herein is a system for sentiment detection in audio data. The system is trained using acoustic information and lexical information to determine a sentiment corresponding to an utterance. In some cases when lexical information is not available, the system (trained on acoustic and lexical information) is configured to determine a sentiment using only acoustic information.

Abstract: Techniques are described for an anomaly detection service for metric data collected by a data monitoring service of a service provider network. The anomaly detection service provides various graphical user interfaces (GUIs), public application programming interfaces (APIs), and other interfaces that enable users to specify metric data of interest to the user and for which the user desires the service to detect occurrences of anomalies. The selected metric data generally can correspond to any type of time series data collected by the data monitoring service and to which a user has access. Example types of metric data that can be monitored by an anomaly detection service include, but are not limited to, operational data generated by various components of a computer system, business data generated by various types of applications, and the like.

Abstract: Techniques are generally described for generation of synthetic image data. In some examples, a selection of a first image may be received. The first image may depict at least a first object having a plurality of image attributes representing visual characteristics of the at least the first object. In some examples, a selection of a first image attribute of the plurality of image attributes to be maintained in subsequently-generated images may be received. In various examples, a first machine learning model may generate a second image having the plurality of image attributes. The change in an appearance of the first image attribute may be minimized in the second image while a change in the appearance of other attributes of the plurality of image attributes may be maximized in the second image.

Abstract: Techniques for improving microphone noise suppression are provided. As wind noise may disproportionately impact a subset of microphones, a method for processing audio data using two adaptive reference algorithm (ARA) paths in parallel is provided. For example, first ARA processing performs noise cancellation using all microphones, while second ARA processing performs noise cancellation using only a portion of the microphones. As the first ARA processing and the second ARA processing are performed in parallel, beam merging can be performed using beams from the first ARA, the second ARA, and/or a combination of each. In addition, beam merging can be performed using beam sections instead of individual beams to further improve performance and reduce attenuation to speech.

Abstract: A system may include a first acoustic event detection (AED) component configured to detect a predetermined set of acoustic events, and include a second AED component configured to detect custom acoustic events that a user configures a device to detect. The first and second AED components are configured to perform task-specific processing, and may receive as input the same acoustic feature data corresponding to audio data that potentially represents occurrence of one or more events. Based on processing by the first and second AED components, a device may output data indicating that one or more acoustic events occurred, where the acoustic events may be a predetermined acoustic event and/or a custom acoustic event.

Abstract: A request to store a container image is received from a device associated with a customer of a computing resource service provider. Validity of a security token associated with the request is authenticated using a cryptographic key maintained as a secret by the computing resource service provider. One or more layers of the container image is built based at least in part on at least one build artifact to form a set of built layers. The software image including the set of built layers is stored in a repository associated with the customer. A manifest of metadata for the set of built layers is stored in a database of a structured data store. The container image is obtained in the form of an obtained container image. The obtained container image is deployed as the software container in at least one virtual machine instance associated with the customer.

Abstract: An emote management system receives a request to make an emote available for users of an application for inclusion in a real-time content stream provided by the application, wherein the emote includes a proposed content pair comprising an image and a text string. The emote management system determines, based on application of a machine learning model to the content pair, whether the proposed content pair satisfies a content pair acceptability threshold. Based a determination that the proposed content pair satisfies the content pair acceptability threshold, return a response to the client indicating that the proposed content pair is an accepted content pair.

Abstract: In accordance with a designation of a private alias endpoint as a routing target for traffic directed to a service from within an isolated virtual network of a provider network, a tunneling intermediary receives a baseline packet generated at a compute instance. The baseline packet indicates a public IP (Internet Protocol) address of the service as the destination, and a private IP address of the compute instance as the source. In accordance with a tunneling protocol, the tunneling intermediary generates an encapsulation packet comprising at least a portion of the baseline packet and a header indicating the isolated virtual network. The encapsulation packet is transmitted to a node of the service.

Abstract: Messages provided by users in a group communication experience are processed to identify media based on content of the messages. The messages may include references to artists, titles, genres, or eras of media, or express sentiments regarding the group communication experience. Media identified based on such references or sentiments is retrieved and transmitted to devices of one or more users participating in the group communication experience. The media is identified for users or for the group communication experience based on subjective or objective factors and played by the devices of the users participating in the group communication experience.

Abstract: Techniques for predicting content availability strategies for content on a content service platform are provided. One technique includes identifying at least a first content item associated with a content provider to make available to one or more users over a communication network. One or more attribute measures of the first content item associated with user interaction with the first content item are determined. An availability framework for the first content item that satisfies one or more predetermined conditions is determined, based on an evaluation of at least the one or more attribute measures. An indication of the availability framework is provided to the content provider.

Abstract: Where a listener consumes media content that is produced on a recurring or non-recurring basis, such as an episode of a media program, the listener may be prompted to receive notifications when the media content, or similar media content, will be aired again in the future. Additionally, where a listener consumes media content while engaged in one or more actions or movements, the media content consumed by the listener is associated with the actions or movements, or a pattern of such actions or movements. Subsequently, when the listener engages in similar actions or movements, or a pattern of such actions or such movements, media content that is similar to the media content previously consumed by the listener during such actions or such movements is identified and recommended to the listener.

Abstract: An electronic book distribution system includes electronic devices that reset their passcodes after specified authentication failures. The passcodes of an individual electronic device is reset to a value that is generated using a predefined function of a randomly generated support code. The support code is displayed to the user, and the user is instructed to contact a support service in order to obtain the new passcode. The support service independently authenticates the user, calculates the new device passcode using the same predefined function used by the electronic device, and provides the new passcode to the user.

Abstract: A logistics system for managing a network of independent delivery partners may include models to detect delivery defects. The models may generate defect scores for each onboarded delivery partner. The system may collect delivery data to continuously update the models. The updated models may score delivery partners for early detection of delivery defects and determine remediations steps. The system may train one or more models to determine the scores based on a terminating step of the delivery steps. The scores may indicate a likelihood the terminating step was caused by intentional action or by accident. The system may determine remediations steps based on the likelihood of intentional action.

Abstract: Devices, systems, and methods are provided for enhanced video annotations using image analysis. A method may include identifying, by a first device, first faces of first video frames, and second faces of second video frames. The method may include determining a first score for the first video frames, the first score indicative of a first number of faces to label, the first number of faces represented by the first video frames, and determining a second score for the second video frames, the second score indicative of a second number of faces to label. The method may include selecting the first video frames for face labeling, and receiving a first face label for the first face. The method may include generating a second face label for the second faces. The method may include sending the first face label and the second face label to a second device for presentation.

Abstract: Systems and methods are described herein for reducing the computational burden related to performing one or more experiments. The set of item assets (e.g., images, text, features, descriptions, etc.) may be reduced in an intelligent manner to enable the set to include more disparate assets. The system may obtain vectors that describe each asset. A similarity score (or other indication/representation of similarity) may be presented for each pair of assets and displayed at a user interface. Using the similarity scores (or similarity representations) as a guide, the user may reduce the set of assets. The reduced set of assets may then be utilized to perform one or more experiments in order to identify an optimal selections from the assets. In some embodiments, the one or more experiments may utilize an explore/exploit algorithm (e.g., a multi-armed bandit algorithm) to identify an optimal selection of item assets.