Abstract: Techniques are disclosed for setting a breakpoint for debugging a neural network. User input is received by a debugger program executable by a host processor indicating a target layer of a neural network at which to halt execution of the neural network. The neural network includes a first set of instructions to be executed by a first execution engine and a second set of instructions to be executed by a second execution engine. A first halt point is set within the first set of instructions and a second halt point is set within the second set of instructions. It is then determined that operation of the first execution engine and the second execution engine has halted. It is then determined that the first execution engine has reached the first halt point. The second execution engine is then caused to move through instructions until reaching the second halt point.

Abstract: A network device can include packet processing circuitry to provide support for virtual functions. The packet processing circuitry can perform operations such as receiving data traffic from a virtual machine, determining an egress rule for the data traffic based on a rule table, and encapsulating the data traffic according to the egress rule.

Abstract: A speech processing system uses contextual data to determine the specific domains, subdomains, and applications appropriate for taking action in response to spoken commands and other utterances. The system can use signals and other contextual data associated with an utterance, such as location signals, content catalog data, data regarding historical usage patterns, data regarding content visually presented on a display screen of a computing device when an utterance was made, other data, or some combination thereof.

Abstract: A system comprising one or more computers implements a synthetic sensor service configured to deploy synthetic sensors to an in-vehicle computing device implementing a synthetic sensor orchestration environment for a vehicle. The synthetic sensor orchestration environment determines a placement decision for a new synthetic sensor to be added to the vehicle based on one or more annotations included in a synthetic sensor package for the new synthetic sensor. The synthetic sensor service and respective synthetic sensor orchestration environments implemented in various types of vehicles provide a consistent way to remotely add additional synthetic sensors and/or other functionality to different types of vehicles after the vehicles have already been put in use by respective owners or operators of the vehicles.

Abstract: A system may include a receiver configured to receive sensor data from one or more aerial vehicles, the sensor data including map data including sensed data related to an aerial view from one or more aerial vehicles of terrain and objects on the terrain. The system may also include a map updating processor in communication with the receiver. The map updating processor may receive the map data and identify a geographic location and/or an orientation associated with the map data. The map updating processor may also align, based at least in part on the geographic location and/or the orientation, the map data with a virtual aerial map providing an aerial view of terrain and objects on the terrain. The map updating processor may also incorporate at least a portion of the map data into the virtual aerial map and generate an updated virtual aerial map.

Abstract: An event deduplication system may efficiently perform event deduplication (identifying “new” or “unique” events that might be an anomaly) by using a first stage that has multiple first stage processes running in parallel (e.g., at different data centers) and a single second stage that has a second stage process that receives and processes events from the different first stage processes. The second stage process updates a global state (e.g., lookup table) and periodically publishes the global state to the first stage processes to update their local state. When the second stage process receives a possible new event from a first stage process, it may more accurately determine whether the event is actually a new event based on the global state.

Abstract: Network pathways are identified to transfer packets between a pair of regional virtual traffic hubs of a provider network. At a first hub of the pair, a first action is performed, resulting in a transmission of a packet received from a first isolated network to the second hub along a pathway selected using dynamic routing parameters. At the second hub, a second action is performed, resulting in the transmission of the packet to a destination within a second isolated network.

Abstract: Combined feature vectors may be generated to map features of two or more search queries to a common embedding space. A user may provide an initial input query and then provide a refinement query. Independent feature vectors may be generated for each of the initial input query and the refinement query, may be weighted, and then may be combined to form a combined feature vector. The combined feature vector aligns different search modalities within the common embedding space that may be executed against an index.

Abstract: A system that determines which applications operating on a device are capable of processing speech processing results. As applications are added or removed, the device may update its own registry information as well as registry information for a remote system, thus ensuring continuity between a local speech processing pipeline and a remote speech processing pipeline. When speech processing is performed, the system can send the results to a corresponding application operating on a local device for further operation, execution, or the like.

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: A materialized view created from one or more shared producer tables in separate producer clusters may be stored in a consumer cluster. The materialized view May undergo faster a refresh using an incremental refresh by obtaining snapshots of the producer table and incrementally refreshing based on deltas obtained by the snapshot. The materialized view may undergo various checks to determine whether the materialized view in the consumer resource is refreshable or unrefreshable based on schema compatibility. The materialized view may undergo further checks to determine whether the materialized view in the consumer resource is available or unavailable for incremental refresh based on detection of schema/data rearrangement operations on the producer table.

Abstract: A first region within image data corresponding to a first portion of a user is determined, and a classifier associated with the first portion of the user is determined. A second region within the image data that includes the first region is determined. Within the second region, first pixels corresponding to the first portion of the user, second pixels corresponding to a second portion of the user adjoined to the first portion of the user, and third pixels corresponding to an environment of the user are identified. Based on identifying the second pixels, the classifier associated with the first portion of the user is confirmed. Upon the confirmation an item is rendered on the first portion of the user.

Abstract: This disclosure describes a system for tracking removal or placement of items at inventory locations with a materials handling facility. In some instances, a user may remove an item from an inventory location and the inventory management system may detect that removal and update a user item list associated with the user to include an item identifier representative of the removed item. Likewise, if the user places an item at an inventory location, the inventory management system may detect that placement and update the user item list to remove an item identifier representative of the placed item.

Abstract: Systems and methods are described for the chained execution of a set of code in an on-demand network code execution system. A user may provide a set of code for execution in the on-demand network code execution system and the system may determine that the set of code comprises multiple chained tasks. The system may provide the set of code to a first virtual machine instance for execution of a first task. The system may obtain an indication that the first task has been executed. The results of the execution of the first task may be sent to a second virtual machine instance, via a push or pull, for execution of a second task. Based on identifying that the first task has been executed, the system may instruct the second virtual machine instance to execute the second task.

Abstract: Described herein are systems and techniques for implementing a third-party item tracking and payment system that enables a user to walk out without a manual checkout process. The techniques include receiving user identifying information associated with a payment account and conveying an identifier of the account to a management system of the retailer. Items, as selected by the user, are added to a virtual cart that is checked out automatically when the user exits the store and the third party system communicates the cart contents to the retailer for cost calculation and then instructs payment using the stored payment information without disrupting existing inventory and account management systems.

Abstract: Techniques are generally described for modeling seasonal relevance for online search. A first query or item comprising text is received. A language model may be used to generate a plurality of token embeddings representing the text. In some cases, the first neural network may predict a seasonal relevance vector for the first query or item based on the plurality of token embeddings, the seasonal relevance vector being predicted based at least in part on a similarity between the text of the first query or item and at least one other query/item for which monthly concentration data is available. The seasonal relevance vector, including a respective seasonal relevance score for each month of a year, may be output.

Abstract: A speech-processing system may determine potential endpoints in a user's speech. Such endpoint prediction may include determining a potential endpoint in a stream of audio data, and may additionally including determining an endpoint score representing a likelihood that the potential endpoint represents an end of speech representing a complete user input. When the potential endpoint has been determined, the system may publish a transcript of speech that preceded the potential endpoint, and send it to downstream components. The system may continue to transcribe audio data and determine additional potential endpoints while the downstream components process the transcript. The downstream components may determine whether the transcript is complete; e.g., represents the entirety of the user input. Final endpoint determinations may be made based on the results of the downstream processing including automatic speech recognition, natural language understanding, etc.

Abstract: Changes made to a database table are accumulated, in durable storage, and snapshots of partitions of the table are obtained. For successive snapshots of a partition, the system accesses a previous snapshot, applies changes from the accumulated changes, and stores the updated snapshot to a durable data store. The accumulated changes and the successive partition snapshots are made available to restore the database to any point in time across a continuum between successive snapshots. Although each partition of the table may have a backup snapshot that was generated at a time different from when other partition snapshots were generated, changes from respective change logs may be selectively log-applied to distinct partitions of a table to generate an on-demand backup of the entire table at common point-in-time across partitions. Point-in-time restores of a table may rely upon a similar process to coalesce partition snapshots that are not aligned in time.

Abstract: On-circuit activity monitoring may be performed to modify integrated circuit processing. An activity monitor may be implemented on an integrated circuit to monitor activity measurements of processing data at another portion of the integrated circuit. A change to activity measurements may be detected and cause the activity monitor to modify the rate at which data enters the other portion of the integrated circuit for processing.