Abstract: Technologies are disclosed for dynamic authorization of users for distributed systems. A permission service allows authorized users to specify permissions for users of different systems, such as physical access security systems (PACS) and easier for users to interact with the authorized resources of the PACS. A manager may select a role for the user to specify the allowed permissions for the user. Changing the permissions/resources of a role automatically changes the permissions for users that are associated with the role. A scope may also be assigned to the user that specifies the locations of the physical resources that the user may access (e.g., a particular data center, a region, globally, . . . ). Instead of having to locate specific resources that the user is authorized to access, a graphical user interface (GUI) displays the authorized resources. When the resources change, the GUI is dynamically updated.

Abstract: An access key retrieval service receives a request from a client device to configure an application on the client device. In response to the request, the access key retrieval service provides a setup code comprising a first component of an authentication key. Additionally, the access key retrieval service provides files for configuring the application, including a manifest file that includes a second component of the authentication key. The client device uses a set of key components that comprises the first component and the second component to derive the authentication key and provides information demonstrating access to the authentication key. The access key retrieval service receives this information and provides an access key usable to enable the application to access computing resources of a service provider.

Abstract: Techniques for synchronized out-of-order live video encoding are described. As one example, a computer-implemented method includes receiving a first real-time encoded live video, that is in a coded order that is different than a display order, at a content delivery service from a contribution encoder at a different location than the content delivery service, performing, in the coded order, a real-time decode of the first real-time encoded live video by a distribution decoder of the content delivery service to generate a decoded video in the coded order, performing, in the coded order, a real-time encode of the decoded video by a distribution encoder of the content delivery service to generate a second real-time encoded video in the coded order without waiting on a reordering of the decoded video into the display order, and transmitting the second real-time encoded video in the coded order to a client device.

Abstract: Systems and methods for performing improper input data detection are described. In one example, a system comprises: hardware circuits configured to receive input data and to perform computations of a neural network based on the input data to generate computation outputs; and an improper input detection circuit configured to: determine a relationship between the computation outputs of the hardware circuits and reference outputs; determine that the input data are improper based on the relationship; and perform an action based on determining that the input data are improper.

Abstract: A biometric identification system uses inputs acquired using different modalities. A model having an intersection branch and an XOR branch is trained to determine an embedding using features present in all modalities (an intersection of modalities), and features that are distinctive to each modality (an XOR of that modality relative to the other modality(s)). During training, a first loss function is used to determine a first loss value with respect to the branches. Probability distributions are determined for the output from the branches, corresponding to the intersection and XORs of each modality. A second loss function uses these probability distributions to determine a second loss value. A total loss function for training the model may be a sum of the first loss and the second loss. Once trained, the model may process query inputs to determine embedding data for comparison with embedding data of a previously enrolled user.

Abstract: This disclosure is directed to techniques in which a first user in an environment scans visual indicia associated with an item, such as a barcode, before handing the item to a second user. One or more computing devices may receive an indication of the scan, retrieve image data of the interaction from a camera within the environment, identify the user that received the item, and update a virtual cart associated with the second user to indicate addition of the item.

Abstract: The present disclosure generally relates to a feedback processing service that can receive customer input, as customer feedback corresponds to a service context. The feedback processing service aggregates semantically similar feedback as a cluster. Then, the feedback processing service can prioritize each of the clusters by ranking each of the clusters.

Abstract: Systems, devices, and methods are provided for secure multiparty computation (MPC) protocols. A first computing entity may send a first cryptographically protected data set to a server and a second computing entity may send a second cryptographically protected data set to the server. The server may lack access to plaintext versions of the data sets. The server may compare cryptographically protected data elements from the first and second data sets as part of a secure MPC protocol to determine certain information regarding the data sets, such as determining which data elements are included in both sets, and perform homomorphic computations according to a homomorphic encryption scheme. The server is accordingly able to determine an encrypted result.

Abstract: This disclosure describes, in part, techniques for identifying user-specific content items and/or time-based content items, as well as techniques for providing the content items to the user. For instance, a remote system may receive first audio data from an electronic device, where the first audio data represents a request to identify content related to a topic. The remote system can then store the topic in a database and use the topic to identify the content. Later, the remote system can then receive second audio data from the electronic device, where the second audio data represents a request for the content. The remote system can then send third audio data to the electronic device that represents at least a portion of the content.

Abstract: Customers of a service provider are able to provision compartments of the accounts. The both the accounts and the compartments, in some embodiments, may have associated computing resources and identities. One or more identities of the account may be authorized to perform administrative operations in the compartment. Identities of the compartment may lack the ability to perform any administrative actions outside of the compartment but inside of the account.

Abstract: Methods and apparatus for conditional master election in a distributed database are described. A plurality of replicas of a database object are stored by a distributed database service. Some types of operations corresponding to client requests directed at the database object are to be coordinated by a master replica. Client access to the database object is enabled prior to election of a master replica. In response to a triggering condition, a particular replica is elected master. The master coordinates implementation of operations with one or more other replicas in response to client requests.

Abstract: Methods and apparatus for identifying consistent snapshot points for file systems in a distributed storage service in which a client may access a file system in backend storage via two or more access nodes. A snapshot coordinator initiates a two-phase commit to perform a snapshot epoch transition. The coordinator sends prepare messages to the nodes. In response, a node sends a response to the coordinator, stops issuing writes to the file system, sets a timer, and waits for a commit message. Upon receiving a response from all nodes, the coordinator sends a commit message. Upon receiving the commit or the timer expiring, the nodes increment an epoch counter and begin issuing writes to the file system tagged with the new epoch and indicating that a copy-on-write operation be performed. The transition is successful if all nodes respond to the request message; otherwise, the two-phase commit is retried.

Abstract: Techniques are disclosed for transferring a message between a sender agent and a receiver agent via a shared memory having a main memory and a cache. Feedback data indicative of a number of read messages in the shared memory is generated by the receiver agent. The feedback data is sent from the receiver agent to the sender agent. A number of unread messages in the shared memory is estimated by the sender agent based on the number of read messages. A threshold for implementing a caching policy is set by the sender agent based on the feedback data. The message is designated as cacheable if the number of unread messages is less than the threshold and as non-cacheable if the number of unread messages is greater than the threshold. The message is written to the shared memory based on the designation.

Abstract: A computing resource service provider may provide computing instances organized into logical groups, such as auto-scaling groups. Computing instances assigned to an auto-scaling group may be place into standby. Standby instances may still be managed by the auto-scaling group but may not contribute to the capacity of the auto-scaling group for auto-scaling purposes.

Abstract: Summaries of media programs that are in progress are generated based on content of the media programs that has already been transmitted to listeners or viewers. The content is transcribed into text, and contextual features regarding the media program such as topics, identities of speakers or interactions received from listeners are identified. The transcribed content and the contextual features are provided as multi-modal inputs to a model that is trained to generate a summary of the media program in response to such inputs. Summaries of media programs that are then in progress are transmitted to devices of listeners who may be interested in joining one of the media programs and displayed in a menu or user interface or announced to the listeners.

Abstract: This disclosure proposes systems and methods for leveraging entity-related language models in speech processing. A system can receive audio data corresponding to an utterance and perform automatic speech recognition (ASR) on a first portion of the audio data using a general language model. Based on the results, the system may identify a specific language model for processing a second portion of the audio data. The specific language model may include entities belonging to a common subject or class. The specific language model may, in some cases, provide better results than the general language model. While the general language model may describe a whole sentence, the specific language model may describe only a portion of a sentence. Thus, a top-level model may “activate” the specific language model when it may provide useful results. The resulting data may include results from both the general language model and the specific language model.

Abstract: Techniques are described for fulfilling an utterance request for an item represented within a video rendered at a client device. In some implementations, a user account associated with the request is identified, enabling a video stream transmitted in association with the user account at the time that the request was uttered to be identified. In one technique, a timestamp associated with the request is used to identify the relevant portion of the video stream. The item represented within the portion of the video stream can be identified using various techniques and/or information such as image recognition, metadata within the video, subtitles, closed captions, and/or a database mapping between the item and a video content item transmitted in the video stream.

Abstract: Features are disclosed for performing functions in response to user requests. Natural Language Understanding (“NLU”) processing may be performed to generate command data that represents a subject of an utterance. The command data may be sent to an application that causes presentation of first output content in a first modality at a first time in response to receiving the command data, and generates second output content in a second modality different from the first modality, wherein the second output content is associated with the first output content. The second output content may be presented in the second modality at a second time subsequent to the first time.

Abstract: Noise that is present in the output of a weight sensor can lead to erroneous weight data. A moveable device, such as a tote, may be used by a customer while shopping in a facility. This tote can include one or more weight sensors that are used to determine the weight of items added to or removed from the tote. However, noise can affect the output of the weight sensors, where such noise is attributed to movement or vibration of the tote. Data from a vibration sensor or a motion sensor coupled to the tote can be analyzed to determine noise that is common to weight data and vibration data or motion data associated with the tote. This common noise can then be removed or attenuated from the weight signals to determine de-noised and valid weight data for the tote.

Abstract: An image processing system may receive image data from a camera of a user device and perform gaze prediction processing of the image data to predict one or more gaze patterns. The gaze prediction processing may include processing the image data using a neural network to detect faces and/or objects and generate an image feature map. The gaze prediction processing may include performing gaze direction prediction operations using the feature map and detected faces and/or objects to determine gaze direction probability data. The gaze prediction processing may include predicting a gaze pattern based on the gaze direction probability data and the image feature map. The gaze pattern may be short-term (e.g., atomic-level) or long-term (e.g., event-level).

Abstract: Systems, methods, and apparatuses are disclosed for external mobility systems for heavy machinery and equipment. In one embodiment, an example system may include a first module, a second module coupled to the first module, and a third module coupled to the first module and the second module. Systems may include an air bearing system disposed under the third module, where the air bearing system is configured to apply an upward force on the third module, and an air supply coupled to the air bearing system. The upward force on the third module may reduce a static friction of the third module with respect to a floor surface by at least 90%, such that the third module can slide from a default position to an expanded position.

Abstract: A first user generates video data for performance of an activity, such as a fitness exercise, by performing the activity in front of a camera. Based on the video data, the amount of movement of different parts of the first user's body is determined. Data representing the position of the first user over time is generated. The data may take the form of a function or a signal that is based on the function. The locations of body parts that move significantly are prioritized over other body parts when determining this data. At a subsequent time, a second user performs the activity. The number of times the second user completes the activity is counted by determining the number of times the second user reaches a position corresponding to a maximum value in the data representing the position of the first user.

Abstract: A data storage system includes a head node and mass storage devices. The head node is configured to store volume data and flush volume data to the mass storage devices. Additionally, the head node is configured to determine a quantity of data partitions and/or parity partitions to store for a chunk of volume data being flushed to the mass storage devices in order to satisfy a durability guarantee. For chunks of data for which complete copies are also stored in an additional data storage system, the head node is configured to reduce the quantity of data partitions and/or parity partitions stored such that required storage space is reduced while still ensuring that the durability guarantee is satisfied.

Abstract: Techniques described herein include systems and methods for utilizing a shopping list generated by a user in one or more shopping list features. A list of items including keywords not associated with specific items may be received. A user interface may be updated to present the list of items, with a first area of the user interface corresponding to uncompleted tasks associated with a keyword of the list, and a second area corresponding to completed tasks associated with a keyword of the list. One or more sensors may receive information about an item. Input about an interaction with the user interface and a keyword of the list may be received. The user interface may be updated to move the keyword of the list from the first area to the second area. A mapping of the keyword to the item may be updated by a computer system.

Abstract: Methods, systems, and computer-readable media for voice-activated selective memory for voice-capturing devices are disclosed. A first voice input from a voice-capturing device is received, via a network, at a service provider environment comprising one or more services. The first voice input comprises one or more utterances from a user of the voice-capturing device. A representation of the first voice input is stored. A second voice input from the voice-capturing device is received, via the network, at the service provider environment. The second voice input represents a command to disregard the first voice input. Based on the second voice input, the representation of the first voice input is deleted.

Abstract: Security policies can be dynamically updated in response to changes in endpoints associated with those policies. A user can indicate one or more regions or networks from which access is to be granted under a specific security policy. The user can subscribe to receive notifications upon a change relating to those endpoints, such as the addition or removal of one or more endpoints. When a change is detected, new policy information can be generated automatically and published for subscribed policies, which can then have the updates applied automatically or provided for manual review and application. Such a process enables access determinations to be made based upon up-to-date endpoint information.

Abstract: In one example, a neural network accelerator can execute a set of instructions to: load a first weight data element from a memory into a systolic array, the first weight data element having first coordinates; extract, from the instructions, information indicating a first subset of input data elements to be obtained from the memory, the first subset being based on a stride of a transposed convolution operation and second coordinates of first weight data element in a rotated array of weight data elements; based on the information, obtain the first subset of input data elements from the memory; load the first subset of input data elements into the systolic array; and control the systolic array to perform first computations based on the first weight data element and the first subset of input data elements to generate output data elements of an array of output data elements.

Abstract: Techniques for performing spoken language understanding (SLU) processing are described. An SLU component may include an audio encoder configured to perform an audio-to-text processing task and an audio-to-NLU processing task. The SLU component may also include a joint decoder configured to perform the audio-to-text processing task, the audio-to-NLU processing task and a text-to-NLU processing task. Input audio data, representing a spoken input, is processed by the audio encoder and the joint decoder to determine NLU data corresponding to the spoken input.

Abstract: Disclosed are various embodiments for using unmanned vehicles to manage radio-based network infrastructure. In one embodiment, a determination is made to add coverage to a radio-based network via a radio unit. An unmanned vehicle carrying the radio unit as a payload is instructed to travel to a particular location. The unmanned vehicle is instructed to deliver the radio unit to the particular location. The radio unit is activated to add the coverage to the radio-based network.

Abstract: Methods and apparatus for partitioned private interconnects to provider networks are described. At least a portion of available bandwidth of a private physical interconnect between a provider network and a connectivity intermediary's network is designated as the bandwidth limit of an interconnect partition set up on behalf of a customer at the request of the intermediary. The intermediary's network comprises one or more devices to which at least one of the customer's devices is connected. Access to one or more resources of the provider network via the interconnect is enabled. Traffic monitoring results associated with the interconnect are used to enforce the designated bandwidth limit of the partition.

Abstract: A system and method for providing contextual recommendations in gaming media streams. One or more users request a gaming media stream broadcast corresponding to game play. A broadcast service obtains the gaming media stream that includes a gaming stream identifier and user identifier. The broadcast service can generate a set of recommendations based on at least one of social network information, historical information, and identification of events and items in the gaming media stream. The broadcast service provides the gaming media stream with one or more recommended items.

Abstract: Technologies are provided for generation of recommendation results using a verbal query. In one embodiment, query data can be generated using a searchable query corresponding to the verbal query. The query data can define a query browse node and a product brand, for example. First product identifiers that match the searchable query can be determined using multiple data repositories. Duplicates from the first product identifiers can be removed, resulting in second product identifiers. Attribute data also can be generated using the second product identifiers. The attribute data can define features for a product identifier. Further, third product identifiers can be determined by applying a filtering model to the second product identifiers. A ranking of the third product identifiers can be generated using an optimization function based on the query data and the product attribute data. A product corresponding to one of the ranked product identifiers represents a recommendation result.

Abstract: An autonomous mobile device (AMD) builds up electrostatic charges from moving and generates heat from the operation of internal components. In addition to possible user discomfort, electrostatic discharges may damage sensors and electronics. Electrostatic charges are dissipated from the AMD using an electrostatic dissipation structure and conductive wheels. A conductive path between a chassis ground, the electrostatic dissipation structure, and the conductive wheels improves the dissipation of electrostatic charges. Electrostatic charges are also dissipated from components by mounting the components using conductive materials. Sensors may be affixed to a support structure that is affected by thermal expansion. Thermal expansion may distort precise positioning of sensors, reducing accuracy of sensor data. An elastomeric foam may be used to mount sensors to a support structure, allowing for thermal expansion without distorting the positioning of the sensors.

Abstract: A hierarchical representation of an input data set comprising similarity scores for respective entity pairs is generated iteratively. In a particular iteration, clusters are obtained from a subset of the iteration's input entity pairs which satisfy a similarity criterion, and then spanning trees are generated for at least some of the clusters. An indication of at least a representative pair of one or more of the clusters is added to the hierarchical representation in the iteration. The hierarchical representation is used to respond to clustering requests.

Abstract: A first worker node of a distributed system computes a first set of gradients using a first neural network model and a first set of weights associated with the first neural network model. The first set of gradients are transmitted from the first worker node to a second worker node of the distributed system. The second worker node computes a first set of synchronized gradients based on the first set of gradients. While the first set of synchronized gradients are being computed, the first worker node computes a second set of gradients using a second neural network model and a second set of weights associated with the second neural network model. The second set of gradients are transmitted from the first worker node to the second worker node. The second worker node computes a second set of synchronized gradients based on the second set of gradients.

Abstract: Systems and methods for sensorless motor control may include a back EMF (electromotive force) observer, an adaptive EMF filter, magnitude compensation, hybrid rotor position and speed determination, rotor position and speed blending, and angle compensation. In order to provide accurate and reliable rotor position and speed estimations for a motor over a wide and varied range of speeds, at low speeds, during speed reversals, and/or in the presence of external forces, loads, or torques, the sensorless motor control may utilize a hybrid rotor position and speed determination that leverages both angle-based and magnitude-based methods. Further, the outputs of the two methods may be blended based on a shaping function to generate a final estimated rotor position and speed. Then, the motor may be more accurately and reliably controlled based on the final estimated rotor position and speed.

Abstract: A model selection and deployment service at a provider network receives an indication of sensor availability from a remote client device (e.g., what type of sensors are currently available to provide sensor data to the client device). The model selection and deployment service then selects, based on the sensor availability (and/or based on one or more other factors/criteria), a data processing model from a group of data processing models that are available for deployment to the client device. The model selection and deployment service then transmits the selected data processing model to the remote client device (e.g., for installation on the hub device). This may allow a client device to use the best data processing model for a sensor configuration and to dynamically adjust to any changes in the sensor configuration.

Abstract: A packet processing technique can include receiving a packet, and parsing the packet based on a protocol field to generate a parse result vector. The parse result vector is used to select between forwarding the packet to a virtual machine executing on a host processing integrated circuit, forwarding the packet to a physical media access controller, multicasting the packet to multiple virtual machines executing on the host processing integrated circuit, and sending the packet to a hypervisor.

Abstract: At a data stream management service, a first set of metadata indicating that a first isolated read channel has been associated with a first data stream is stored. The first isolated read channel has an associated read performance limit setting. A second set of metadata indicating that a second isolated read channel, with its own performance limit setting, has been associated with a data stream is also stored. Based on determining that the difference between a metric of read operations associated with the first channel and the read performance limit setting of the first channel meets a first criterion, the service initiates a throttling operation for reads associated with the first channel. The throttling decision is made independently of read metrics of the second channel.

Abstract: Embodiments describe a method for decoding speech including receiving speech input at an audio input device, generating speech data that is a digital representation of the speech input; extracting acoustic features of the speech data, assigning acoustic scores to the acoustic features, receiving data representing the acoustic features and the acoustic scores, decoding the data representing the acoustic features into a word, having a word score, by referencing a WFST language model, modifying the word score into a new word score based on a personalized grammar model stored in the external memory device, the processor is separate from and external to the WFST accelerator, and determining an intent represented by a plurality of words outputted by the WFST accelerator, where the plurality of words include the word and the new word score.

Abstract: Techniques for intent-based access control are described. A method of intent-based access control may include receiving, via a user interface of an intent-based governance service, one or more intent statements associated with user resources in a provider network, the one or more intent statements expressing at least one type of action allowed to be performed on the user resources, compiling the one or more intent statements into at least one access control policy, and associating the at least one access control policy with the user resources.

Abstract: A speech interface device is configured to receive response data from a remote speech processing system for responding to user speech. This response data may be enhanced with information such as remote NLU data. The response data from the remote speech processing system may be compared to local NLU data to improve a speech processing model on the device. Thus, the device may perform supervised on-device learning based on the remote NLU data. The device may determine differences between the updated speech processing model and an original speech processing model received from the remote system and may send data indicating these differences to the remote system. The remote system may aggregate data received from a plurality of devices and may generate an improved speech processing model.

Abstract: Deployable navigation beacons can be deployed from a vehicle, such as an unmanned aerial vehicle (UAV), in an event of a loss of position or orientation of the vehicle. After deployment of the navigation beacons, the vehicle may detect locations of the navigation beacon, which may define a surface that may include surface features. The vehicle may then perform control operations based on the resolved locations. For example, UAV may maneuver to land proximate to the navigation beacons after resolving locations of the navigation beacons as a continuous surface. The navigation beacons may output a visual signal (e.g., a light), a auditory signal (e.g., a sound), and/or a radio signal. In some embodiments, each navigation beacon may include a different or unique signal.

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.