Abstract: Techniques for generating, from first speaker recognition data corresponding to at least a first word, second speaker recognition data corresponding to at least a second word are described. During a speaker recognition enrollment process, a device receives audio data corresponding to one or more prompted spoken inputs comprising the at least first word. Using the prompted spoken input(s), the first speaker recognition data (specific to that least first word) is generated. Sometime thereafter, a user may indicate that speaker recognition processing is to be performed using at least a second word. Rather than have the user go through the speaker recognition enrollment process a second time, the device (or a system) may apply a transformation model to the first speaker recognition data to generate second speaker recognition data specific to the at least second word.

Abstract: A fiber optic connector can include a connector body with a connector channel extending therethrough. A ferrule can be positioned in a first end of the connector channel and a hollow core fiber can be positioned in a second end of the connector channel. A lens can be positioned in the connector channel between the ferrule and the hollow core fiber. The lens can direct light emitted from the hollow core fiber to an optical fiber within the ferrule.

Abstract: Techniques for filtering the output of supplemental content are described. When a supplemental output system (e.g., a supplemental content system or notification system) receives supplemental content for output, the supplemental output system sends a user identifier (of the recipient user) and the supplemental content to separately implemented filtering component. The filtering component uses a machine learning (ML) model to determine a topic of the supplemental content. The filtering component determines whether the supplemental content should not be output based on the ML model-determined topic, one or more guardrail policies of the supplemental output system, and user frustration data regarding previously output supplemental content. Use of the ML model to determine the topic prevents a content publisher from surreptitiously associating supplemental content with a specific topic in an effort to bypass topic-based output guardrails.

Abstract: An electronic device includes a first passive infrared (PIR) sensor and a second PIR sensor positive vertically below the first PIR sensor. One or more lenses are shaped and positioned to focus light received at the one or more lenses onto the first PIR sensor and the second PIR sensor. The first PIR sensor has a first field of view (FoV), and the second PIR sensor has a second FoV that is vertically and horizontally offset from the first FoV.

Abstract: Domain specialty instructions may be generated for performing text analysis tasks. An input text may be received for performing a text analysis task. One or more domain entities may be extracted from the input text using a machine learning model trained to recognize entities of a domain in a given text. The one or more domain entities may be inserted as part of generating instructions to perform the text analysis task using a pre-trained machine learning model fine-tuned to the domain. The pre-trained machine learning model may be caused to perform the text analysis task using the generated instructions and a result of the text analysis task may be provided.

Abstract: A computing system that provides virtual computing services may generate and manage remote computing sessions between client computing devices and virtual desktop instances hosted on the service provider's network. A computing resource instance manager may monitor connections to and disconnections from a virtual desktop instance during particular time periods, and may apply a resource management policy to determine whether and when to shut down an underlying virtualized computing resource instance following a disconnection (e.g., immediately, after some period of time, or only between certain hours). A storage volume for the virtual desktop instance may be detached during a shutdown. In response to a reconnection request, the virtualized computing resource instance (or another such instance) may be restarted and the storage volume may be reattached. The computing resource instance manager may develop a model for predicting when to shut down or restart an instance based on historical data or machine learning.

Abstract: Pairs of text collections are obtained. An individual pair comprises (a) a source text collection which includes a first group of text sequences and (b) an annotated analysis result of the source text collection, comprising a second group of text sequences and a set of evidence mappings generated by an evidence mapping model. An evidence mapping indicates, for a particular text sequence of the second group, another text sequence of the first group which provides evidence for the particular text sequence. A quality metric of the model is obtained using an automated evaluation methodology in which a question is generated from the particular text sequence, and an analysis of a pair of answers (including an answer generated using an evidence mapping) to the question is performed. The quality metric is provided via a programmatic interface.

Abstract: Pairs of text collections are obtained. An individual pair comprises (a) a source text collection which includes a first group of text sequences and (b) an annotated analysis result of the source text collection, comprising a second group of text sequences and a set of evidence mappings generated by an evidence mapping model. An evidence mapping indicates, for a particular text sequence of the second group, another text sequence of the first group which provides evidence for the particular text sequence. A quality metric of the model is obtained using an automated evaluation methodology in which a question is generated from the particular text sequence, and an analysis of a pair of answers (including 10 an answer generated using an evidence mapping) to the question is performed. The quality metric is provided via a programmatic interface.

Abstract: Systems and methods for securing optical fibers with complimentary tapered ends are described. In some embodiments, optical fibers are aligned to form an adiabatic coupling, and a first one of the optical fibers is secured, via an adhesive, to a structure of an optical device hosting a second one of the optical fibers with which the first optical fibers is coupled. The first and second optical fibers, coupled and secured using the adhesive, are then immersed in a photo-active liquid polymer and two-photon lithography is used to form an additional securing structure around the adiabatically coupled ends of the first and second optical fibers. The additional securing structure is configured to maintain the adiabatic coupling throughout various mechanical and/or thermal shocks the adiabatically coupled fibers may encounter.

Abstract: A provider network implements service discovery and renaming. A provider network includes a domain name service (DNS) and a renaming tool. The DNS receives queries for service names and sends query logs to a data store. Each query log indicates the queried name and a network address (e.g., IP address). A user may request that the service be renamed to a new name. In response, the renaming tool adds a new entry to for the new name to the DNS. The renaming tool retrieves data from the query logs and identifies clients of the service. The renaming tool sends a notification to the identified clients that the name has changed to the new name. The renaming manger retrieves additional data from the query logs to identify how many of the clients have switched to using the new name. After enough clients have switched, the old name may be decommissioned.

Abstract: An object storage system includes mass storage devices that implement general storage for objects stored in the object storage system and additionally includes other storage devices, such as solid-state drives, that provide higher performance storage access. The object storage system implements a common access interface for accessing both accelerated access objects (who are eligible to have cached copies stored on the higher performance storage devices) and non-accelerated access objects stored in the general storage. The cache is fully managed by the service and no changes are required for client applications to receive accelerated access to objects that are classified as accelerated access objects per a customer configurable acceleration policy for the object or for a bucket in which the object is stored.

Abstract: A system and method for distributing quantum entanglement using optically protected fiber optical paths is described. In some embodiments, multiple optical paths are used to connect a source site to a receiver site. Quantum entangled particles are transmitted from the source site to the receiver site using a first one of the multiple optical paths and related service channel information is transmitted from the source site to the receiver site using a second one of the multiple optical paths. In response to a failure of either the first or second optical path, an entanglement distribution controller automatically updates the routing such that the service channel information and the quantum entangled particles are routed concurrently on a remaining one of the first or second optical path.

Abstract: Techniques for calculating and using a request cost index for throttling application programming interface (API) requests to execute operations in a provider network. The techniques encompass the step receiving a request to execute an operation at an API service in the provider network. Further steps include determining to execute the operation based on a request rate limiting algorithm, executing the operation to yield an operation result, and sending the operation result. Additional steps include calculating a request cost index that reflects an amount of computing resources utilized by executing the operation, determining an adjustment amount for a state variable of the request rate limiting algorithm based on the calculated request cost index, and adjusting (e.g., lowering) the state variable by the adjustment amount. Other (e.g., subsequent) requests to execute queries received at the API service that are metered by the state variable can be throttled by the API service.

Abstract: Techniques are described for performing actions (e.g., streaming video content, such as advertising content) based on visual markers detected within a live video feed being streamed to one or more computing devices. First video content is streamed to the one or more computing devices. A presence of a content item within the first video content is detected at a first point in time during streaming of the first video content. An action associated with the content item is performed at a second point in time in response to detection of the presence of the content item in the first video content.

Abstract: A system configured to perform content identification using fingerprinting to recognize known media content. The system may generate a reference database including reference fingerprints for each media content item to include in the content identification. In addition, the system may generate a hash table that associates individual frames of the reference fingerprints with identification information for corresponding media content items. When a device is playing media content, the system may perform content identification by generating query fingerprints representing the media content and comparing the query fingerprints to the reference database. For example, the system may match a query fingerprint to a reference fingerprint by identifying which of the reference fingerprints shares the most frames with the query fingerprint using the hash table. In addition, the system may use additional decision criteria to confirm a match, such as fine-grain matching or tracking successive fingerprints over time.

Abstract: This disclosure describes, in part, techniques for adapting bandwidth estimation algorithms to account for local network stalls (e.g., Wi-Fi stalls). For instance, an electronic device may use a pacer component and/or a network socket buffer to detect Wi-Fi stalls. The pacer component and/or the network socket buffer may both be configured to store packets. The pacer component may determine a number of bytes to send to the network socket buffer based on a current bandwidth (or bitrate) estimation value and may move packets from the pacer component to the network socket buffer. If the second queue reaches capacity at a first time and is no longer at capacity at a second subsequent time, a Wi-Fi stall may be detected, and the electronic device may transition from a first state to a second state causing the electronic device to determine whether to update a current bandwidth estimation value.

Abstract: Machine learning models may be generated that are compatible with another machine learning model and satisfy a resource constraint. Techniques that ensure weight compatibility and architectural compatibility between a machine learning model being created to be compatible with another machine learning model are applied. The resource constraint is enforced so that the generated machine learning model also fits within the resource constraint.

Abstract: Techniques for utilizing machine learning to generate and use publishing parameters for a content delivery system are described.

Abstract: Systems, devices, and methods are provided for package delivery guidance and assistance. A vehicle may comprise one or more sensors. A processing unit of the vehicle may determine, based on vehicle data obtained from the one or more sensors, that the vehicle is in a first state, determine that the first state corresponds to a first sub-task of the plurality of sub-tasks, determine a first graphical interface for performance of the first sub-task, present the first graphical interface on a first display screen of the at vehicle, responsive to additional vehicle data obtained from the one or more sensors indicating that the vehicle is in a second state: determine that the second state corresponds to a second sub-task of the plurality of sub-tasks, and update the first display screen with a second graphical interface that is for performance of the second sub-task.

Abstract: A systolic array can implement an architecture tailored to perform matrix multiplications on sparse matrices. Each processing element in the systolic array may include a register configured to store a value, and a multiplexor configured to select an input element from multiple input data buses based on metadata associated with the value. Each processing element may also include a multiplier configured to multiply the selected input element with the value to generate a multiplication result, and an adder configured to add the multiplication result to a partial sum input to generate a partial sum output.