Abstract: A vector-labeled database may improve a cache hit rate for read requests by performing a partial approximate nearest neighbor index search at a router to designate the cache that results of a read request are to be written to. The selected cache may be associated with a particular node of a layer of an approximate nearest neighbor index search. The router may direct all reads with search vectors close to the particular node to the associated cache.

Abstract: Selective frequency dissipation is implemented that enables cooling of energy gap protected qubits such that excited energy states resulting from heating or other undesired processes are returned to a lower excited energy state or a ground state manifold, thus reducing the probability of errors. Also, the selective frequency dissipation inhibits leakage from the energy gap protected qubits when in the ground state. Additionally or alternatively, Hamiltonian engineering by inducing parity rotations is implemented to decouple the wells of the energy gap protected qubit to further reduce errors when leakage does happen.

Abstract: A vector-labeled database, which stores data in association with multi-dimensional vectors, comprises a hot tier which is optimized for writes, a cold tier which is optimized for reads, and a medium tier which is optimized to balance reads and writes. Incoming transactions are executed at the hot tier, and moved to the medium and cold tier in batches. Reads for the vector-labeled database search all tiers, and merge the results of the searches of the individual tiers.

Abstract: Devices and techniques are generally described for machine translation. In various examples, first text in a first spoken language may be received. A first candidate translation of the first text in a second spoken language may be determined based at least in part on the first text from among a candidate pool of non-translated text associated with a first domain. Combined data representing the first text and the first candidate translation may be generated. A first machine learning model may generate second text by modifying at least one word of the first candidate translation. The second text representing a machine translation of the first text may be output.

Abstract: Techniques for filtering for harmful generative artificial intelligence (AI) results are described. An example of filtering includes receiving a request for an input phrase to be responded to by a generative AI model; comparing to the received input phrase to at least one known harmful input phrase to determine that the received input phrase is to be provided to the generative AI model; providing the received input phrase to the generative AI model; and generating a response by at least in part on an output of the generative AI model.

Abstract: A station is used to induct item(s) into an environment. The station includes a case module configured to receive the item(s), which may be packaged in a box, and sensor(s) configured to scan the box. A camera and lighting module of the station may read the item(s) as they are unpackaged from the box and transferred into one or more totes located on a tote module. The item(s) may be passed over the camera and lighting module and then placed into the tote. Sensor(s) are used to determine which of the one or more totes the item(s) are placed. As the tote(s) become full, the tote(s) may be inducted into the environment for storage, order fulfillment, etc. The station may be used to conveniently remove item(s) from their boxes and place the item(s) into the totes that allow automated storage systems to move the totes throughout the environment.

Abstract: Approaches presented herein relate to the management of secure secrets in a distributed environment. In particular, various embodiments provide for the management of unique digital identities across multiple regions, where each region can include its own certificate authority. While these certificate authorities may operate independently, they can be part of a multi-primary system where unique identities and keys are stored redundantly across environments. In the event of a failure of a certificate authority in one region, another certificate authority in another region can continue security and authentication management, without a need to issue new identities or change operation of any of the regions. Parties to secure communications, such as application containers, can each receive their own unique identity which can be shared across various regions to allow related tasks (e.g., certificate issuance or revocation) to be performed identically from any of those regions.

Abstract: Systems and techniques for picking an item from a container using an end effector are described. An example system includes a rack comprising a plurality of containers. Each container is configured to hold one or more items. The system also includes a robotic arm having an end effector configured to pick a target item from one of the plurality of containers. The end effector includes (i) first and second plates disposed parallel to each other and (ii) an extendable suction cup disposed between the first and second plates. The end effector is configured to extend the extendable suction cup to contact the target item, create a seal with the target item, after creating the seal, retract the extendable suction cup to remove the target item from the container, and after retracting the extendable suction cup, engage the target item between the first and second plates.

Abstract: Techniques for heat management in a serverless managed database service in a cloud provider network are described. The service receives a request to launch a database instance that can scale resource usage based on a computing resource demand of a database application. A set of candidate host computer systems to host the database instance is identified. A host resource usage metric of each host computer system in the set of candidate host computer systems is obtained. A host computer system from the set of candidate host computer systems is selected to host the database instance, the selection based at least in part on the host resource usage metric of the selected host computer system. The database instance is launched on the selected host computer system.

Abstract: Provided is a system for facilitating continuous integration and continuous deployment of data warehousing workflows in a cloud network environment. A workflow integration and deployment service may communicate with a code repository service to be notified of code changes, obtain code changes, generate updated workflows, test updated workflows, and deploy updated workflows. The workflow integration and deployment service may maintain workflow generation and testing rules to generate, test, and deploy workflows according to user preferences.

Abstract: Systems and methods are provided for an improved business intelligence platform that maintains and uses personalized data cache for processing natural language queries with personalized fuzzy matching in lieu of using predefined indexes of large amounts of data. The systems and methods provide for performing fuzzy matching using a natural language query and a personalized data cache storing a copy of data previously provided in response to interactions associated with a user profile, wherein the personalized data cache is separate from a data source storing the response data, and wherein the fuzzy matching does not use any fuzzy index of the data source generated prior to receiving the natural language query; obtaining, from results of the fuzzy matching, a set of query terms; obtaining the response data from the data source using the set of query terms; and generating a response to the natural language query using the response data.

Abstract: Techniques for securing DHCP communications between DHCP servers and client devices in a network to authenticate the devices before providing IP addresses. A client device may store one-time programmable (OTP) seed parameters, and use the OTP seed parameters to derive a unique private key. The private key is used to sign payloads of each DHCP packet sent to the DHCP server. The DHCP server is provided with a public key that verifies signatures made using the private key, and is also provided a list of unique identifiers for client devices that are permitted into the network. The DHCP server uses the public key to verify signatures on DHCP packets, and further verifies that the unique identifier in the DHCP packets is included in the list of expected unique identifiers. Firmware code measurements can be sent in the DHCP packets for the DHCP server to authenticate firmware of the client device.

Abstract: Systems and methods are disclosed for implementation of a scalable and efficient scaling algorithm for clustering data objects within a data set, as well as implementation of the algorithm to control content generation and selection. A data set can be initially divided randomly into subsets. A base clustering algorithm, such as k-means clustering, can be applied in parallel to each subset to identify a set of local centroids representative of local clusters within each subset. The local centroids can be aggregated from across subsets and the base clustering algorithm can be applied to the aggregated local centroids to result in global clusters. Data objects associated with each global cluster (such as by being clustered within a local cluster corresponding to a global cluster) represent a global clustering of the data set. These global clusters can form the basis for additional iterations of local and global clustering, further refining global clustering.

Abstract: A system receives a request for content insertion in a web page for presentation at a device. The request includes contextual information about web traffic between the device and a web server, and the request excludes an identifier of a user of the device. The system associates the request with a cohort that represents a set of previous content insertion requests and characterized by a set of attribute values. The system determines a property of the cohort. The property indicates a count of previously presented content or a decision regarding blocking a future insertion of the content. The system determines that the content can be inserted in the web page. The system sends to the web server in response to the request, an indication to insert the content in the web page. The indication causes the content to be shown in the web page upon presentation at the device.

Abstract: Centralized data set management techniques are implemented for data sets for machine learning applications. New data sets may be created from a search of managed data set labels. Data objects of the managed data sets that are found in the search are then used to generate a new data set. The new data set is then made available for machine learning applications to use.

Abstract: Techniques for permission controls from a first provider for a second provider are described. In an example, a computer system stores first account data of a user with a first service provider. The first account data enables execution of a first online service of the first service provider to provide a second online service of a second service provider. The computer system receives second account data of the user with the second service provider. The computer system generates, based on a mapping between a first schema of the second account data and a second schema, third account data. The computer system generates, based on the first account data and the third account data, a report associated with controlling execution of the second online service. The computer system sends permission data to the second service provider indicating whether a permission exists to control the execution of the second online service.

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: A system including a one or more de-jamming mechanisms having a wedge configured to move from a retracted state to an extended state in which the wedge is at least partially disposed within a material handling equipment (MHE) in which packages are processed. The system receives data associated with the packages in the MHE, and causes, at a first instance in time, the wedge to move from the retracted state to the extended state, and causes, at a second instance in time, the wedge to move from the extended state to the retracted state.

Abstract: Described are apparatuses, assemblies, and processes to create a carbon fiber part with a smooth convex surface using an interior mold. An uncured part may be formed by wrapping fiber and resin around a mandrel (interior mold). One or more caul plate and optional shims may be placed over the uncured part in alignment with features that correspond between the caul plate(s) and the mandrel. The assembly may be enclosed in a vacuum bag and consolidated. The assembly may be placed on a jig base. A compression member may be coupled to the jig base to contain the assembly therebetween. The compression member may exert a force onto the caul plate to compress the uncured part against the mandrel during a curing process, thereby creating a smooth surface finish of the resulting cured part that is free from substantial undesirable warping or wrinkles about the surface.

Abstract: Techniques for generating a multi-modal, context aware embedded representation of an entity and contextual data associated with the entity are described. The system may receive a natural language entity for which an embedded representation is to be generated. The system uses the natural language entity to determine/generate various context data associated with the entity (e.g., slot type, usage example, phonemes, audio data, image/video data, etc.). The system may tokenize and encode the natural language entity and the context data. Thereafter, the encoded entity and the context data may be inputted to a language model, which generates a single embedded representation corresponding to the entity and the context data. Thereafter, the embedded representation may be used by a system component for performing processing with respect to a user input.

Abstract: Adaptive code enhancement includes receiving feature data of a speech waveform having a plurality of frames, inferring a bitrate based on a quantity of bits of the plurality of frames received from the decoder codec, generating one or more latent feature vectors in accordance with the bitrate and based on the feature data, receiving one or more pitch lags and a speech signal, calculating filter coefficients on a per-frame basis based on the one or more latent feature vectors and the one or more pitch lags, and modifying the speech signal using the one or more filter coefficients to generate a modified speech signal.

Abstract: Systems and methods for managing a representation of a transactional table of a transactional database in an analytical database in near real-time are disclosed. A historical mode enables the analytical database to maintain both a current representation of the transactional table, but also one or more historical representations. The table representation maintained by the analytical database can enter and exit the historical mode without a need for re-start. For example, additional columns specifying whether a given row is active and what transaction numbers it was active for, are added to allow the historical representations to be included in the same table with the current representation.

Abstract: Frames may be encoded at different resolutions. A scaling factor may be used to scale video frames used for translational or affine motion compensation. Various improvements to motion compensation processes for different resolution video frames are described.

Abstract: Techniques for generating a manifest, for a multimedia file, that utilizes a hierarchical data structure with (i) nodes for the manifest request parameters and (ii) the instructions for how to generate a manifest based on that manifest request parameter included within the node are described.

Abstract: Technology directed to a radar detection filtering approach is described. Processing logic obtains radar data from a radar unit. The processing logic determines, based on the radar data, that a candidate detection event has occurred. The processing logic determines signal characteristics of the candidate detection event. A confidence score is generated using the signal characteristics. The confidence score is compared to a confidence threshold, and in response to the confidence score satisfying the confidence threshold, a process to capture image(s) or video is activated.

Abstract: A hybrid serverless computing and durable computing search system provides low-latency search by maintaining domain specific search indices in memory or storage, while also allowing for low-cost service by scaling to zero when not in use. An initial search query is processed by an ephemeral computing instance loaded with code for executing search and which has downloaded relevant search indexes for a given domain. Asynchronously, a longer-lived container is launched and loaded with the relevant search indices. Subsequent search queries for the given domain are handled using the container without the delay associated with downloading the relevant search indices. After a period of inactivity both the container and ephemeral computing instances are scaled to zero.

Abstract: A distributed computing system comprises a plurality of clusters of computing devices, including at least a first cluster with a first number of computing devices and a second cluster with a second number of computing devices. In response to receiving a request to process a query, the distributed computing system analyzes the query to generate an estimate of computing resources to be used when processing the query. The distributed computing system selects between the first and second cluster, based at least in part on the estimate, and tentatively assigns the query to the selected cluster for processing. The cluster to which the request is assigned determines if it has sufficient capacity to process the request.

Abstract: Disclosed are systems and methods to intelligently present user interfaces that are dynamically generated based on the determined intent or needs of the user. For example, based on user requests or actions, the disclosed implementations may dynamically generate and present to a user one or more components that are relevant to the user. Such components may be generated and presented anywhere in a provider network environment, from within a console or home page to guide a user to a service and/or to respond to a user query, to within a software development service to provide a developer with code or other requested content.

Abstract: Techniques for providing graphic overlays on live videos using a machine learning (ML) model are described.

Abstract: A network protocol provides mutual authentication of network-connected devices that are parties to a communication channel in environments where the amount of memory and processing power available to the network-connected devices is constrained. When a new device is added to a network, the device contacts a registration service and provides authentication information that proves the authenticity of the device. After verifying the authenticity of the device, the registration service generates a token that can be used to by the device to authenticate with other network entities, and provides the token to the device. The registration service publishes the token using a directory service. When the device connects to another network entity, the device provides the token to the other network entity, and the other network entity authenticates the device by verifying the token using the directory service.

Abstract: Systems and methods are described for facilitating cross-account network traffic between serverless function executions and isolated cloud resources within virtualized network environments. Virtualized network environments, by default, may be isolated such that external traffic is not permitted to enter the environment. Permissions for traffic that may enter the environment are often set on the basis of network addresses. In the context of serverless functions, such permissions may be difficult to establish because executions of serverless functions can occur on a dynamically selected environment without a fixed network address. The present disclosure provides for creation of user-defined connectors that facilitate routing of network traffic between executions of serverless functions and user virtualized network environments without requiring that routing occur on the bases of network addresses.

Abstract: Voice-controlled devices that include one or more speakers for outputting audio. In some instances, the device includes at least one speaker within a cylindrical housing, with the speaker aimed or pointed away from a microphone coupled to the housing. For instance, if the microphone resides at or near the top of the cylindrical housing, then the speaker may point downwards along the longitudinal axis of the housing and away from the microphone. By pointing the speaker away from the microphone, the microphone will receive less sound from the speaker than if the speaker were pointed toward the microphone). Because the voice-controlled device may perform speech recognition on audio signals generated by the microphone, less sound from the speaker represented in the audio signal may result in more accurate speech recognition, and/or a lesser need to perform acoustic echo cancelation (AEC) on the generated audio signals.

Abstract: Virtual resources may be provisioned in a manner that is aware of, and respects, underlying implementation resource boundaries. A customer of the virtual resource provider may specify that particular virtual resources are to be implemented with implementation resources that are dedicated to the customer. Dedicating an implementation resource to a particular customer of a virtual resource provider may establish one or more information barriers between the particular customer and other customers of the virtual resource provider. Implementation resources may require transition procedures, including custom transition procedures, to enter and exit dedicated implementation resource pools. Costs corresponding to active and inactive implementation resources in a dedicated pools associated with a particular customer may be accounted for, and presented to, the customer in a variety of ways including explicit, adjusted per customer and adjusted per type of virtual resource and/or implementation resource.

Abstract: Devices and techniques are generally described for entity management. In various examples, a natural language request comprising a reference to an entity may be received. Resolved entity data associated with the entity may be determined. Identifier data associated with the resolved entity data may be generated and the resolved entity data may be stored in association with the identifier data in a session cache. A first prompt that comprises the identifier data may be generated. A language model (LM) may generate a data request that includes the identifier data as an input parameter. At least a portion of the resolved entity data may be received from the session cache using the data request. A second prompt that comprises at least the portion of the resolved entity data may be generated and used to perform an action with respect to the entity in response to the natural language request.

Abstract: Systems and methods for visual event processing using multi-modal language models include receiving, at a first device, first user input data representing a first user command to store occurrence of a type of visual, auditory or other event over time, the event associated with a physical activity involving movement, sound, or other type of detectable physical stimuli. An activity indicator describing the activity may be stored in the first device. The first device's functionality can be extended to process data from the sensors of a second device. As such, the activity indicator stored in the first device can be used to process a first query requesting that a model identify the activity from the data. The model may determine that the data depicts the activity and a first output (natural language, notification, smart home device response, etc.) may be generated indicating occurrence of the event.

Abstract: Approaches for determining continuous confidence scores for rule-based negation detection systems overcome deficiencies of existing rule-based negation detection systems that provide only binary output (e.g., negated or not negated). The approaches encompass a pointwise mutual information (PMI)-based approach in which PMI is measured between negation cue words and gold-standard negation annotations in a reference dataset. The approaches also encompass a conditional probability-based approach in which the condition probabilities that gold-standard negations exist in a reference dataset for negation cue words are measured. In both approaches, a form of association between negation cue words and gold-standard negations are measured to provide a more accurate indication of the rule-based system's confidence in a negation determination compared to a binary output.

Abstract: The present disclosure generally relates to systems and methods for generating and evaluating a transcript summary based on a transcript of an active meeting. A summary generation system may generate a summary based on an active meeting while minimizing latency and the number of calls to a large language model. The summary generation system may provide each transcript chunk of a transcript as input into an LLM for summarization and append to a transcript summary. In addition, the summary generation system may determine whether the transcript summary of the transcript has reached a threshold, such as a length or word count threshold. If the threshold has been reached, the summary generation system may input the transcript summary into the LLM to generate a revised transcript summary. The revised transcript summary may replace the existing transcript summary.

Abstract: A technique for rematerializing a tensor in a neural network model may include identifying a candidate operator for rematerialization in a representation of a data flow graph for a neural network model. An operator can be identified as a candidate operator if it generates a tensor that is used multiple times in the neural network model and belongs to a set of permitted operators for rematerialization. An operator chain that starts at one or more source operators and ends at the candidate operator is determined. A profitability of rematerializing the operator chain is determined based on a comparison of a memory footprint of source data for the one or more source operators, and a memory footprint of the tensor. If it is determined that rematerializing the operator chain is profitable, the operator chain is inserted in the representation of the data flow graph where the tensor is reused.

Abstract: In various examples, systems and methods for robotic retrieval are described. A storage block including a grid of a plurality of storage pods in a robot-controlled storage field may be determined. Each storage pod of the plurality of storage pods may include a plurality of items. The grid may include a plurality of cells, where each cell represents a position within the grid occupiable by a storage pod. A first storage pod located at a first cell of the grid may be determined. The first cell may be an interior cell of the grid. A second cell of the grid may be determined, where the second cell is unoccupied by a storage pod. At least one robotic drive may be controlled to shift storage pods toward the second cell.

Abstract: Techniques to improve the reliability of volatile memory may include receiving a write request to store a data chunk in a volatile memory implemented using a memory module that contains a plurality of memory sub-channels. In response to the write request, a first write access is performed to write at least a first portion of the data chunk into a first memory sub-channel of the memory module, and a second write access is performed to write at least a second portion of the data chunk into a second memory sub-channel of the memory module.

Abstract: An operating system, or operating system update, management service uses a shared read-only multi-attach volume of a block-based storage service to distribute operating systems or operating system updates to a set of virtualized computing instances. Also, to store launch specific information, that is specific to a given virtualized computing instance, additional writable volumes are used, wherein a write volume is attached to each of the computing instances of the set. This eliminates the need to provide a full copy of an OS volume to each of the computing instances.

Abstract: Techniques described herein relate to swing anomaly detection for objects transferred via robotic manipulation. For example, a computer system can generate, based on first sensor data of a robotic manipulator manipulating an object and by using a machine learning model, a first segmentation of the object represented by the first sensor data. The first sensor data can be received at a first time. The computer system can generate, based on second sensor data of the robotic manipulator manipulating the object and by using the machine learning model, a second segmentation of the object represented by the second sensor data. The computer system can generate, by comparing the first segmentation and the second segmentation, a swinging score indicating swinging of the object by the robotic manipulator. A motion plan for the robotic manipulator can be updated based on the swinging score.

Abstract: A system that improves beam cancellation by refining reference beam selection and dynamically controlling an adaptation speed of adaptive filters. For example, a device may control the adaptation speed by dynamically determining a variable step-size parameter based on a relative strength of a microphone signal and/or a signal-to-noise ratio (SNR) value associated with an individual beam. The device may compare current energy levels of the microphone signal to a range of energy levels to determine a microphone step-size value, track a minimum noise floor to determine an SNR step-size value, and determine the variable step-size parameter using a sigmoid curve. Additionally or alternatively, the device can use a hybrid approach to select reference beams and can perform neighbor exclusion to further protect speech. For example, the device may exclude reference beams that are adjacent to a target beam to prevent the speech from being represented in the reference beams.

Abstract: A capacitive mat includes a first set of conductors that function as sensing electrodes applied to a first side of a substrate and a second set of conductors that function as shield electrodes applied to a second side of the substrate. Upper and lower exterior layers are placed over the sets of conductors. The substrate and the exterior layers may be formed from flexible materials. Circuitry in the mat determines a capacitance associated with the conductors at a region of the mat that is associated with an object on the mat. To reduce thickness and improve flexibility, the shield electrodes may be used to determine a ground reference value and use of separate ground electrodes may be omitted. Conductive regions on the mat may be used to couple with adjacent mats, allowing several mats to be easily combined to cover a larger surface.

Abstract: Disclosed are various embodiments for dynamic spectrum management for portable radio-based networks. In one embodiment, it is determined that a portable radio unit implementing a cell of a radio-based network has moved away from a reference location associated with a first spectrum allocation used by the portable radio unit. A second spectrum allocation for the portable radio unit is obtained based at least in part on a direction of movement of the portable radio unit. The portable radio unit is then configured to use the second spectrum allocation instead of the first spectrum allocation.

Abstract: Video scene complexity can be estimated at a camera in real time. The camera may store associations between video complexity score ranges and corresponding bitrate values and between maximum key frame sizes and corresponding compression parameter values. As video is captured, a compression parameter value, a size of a compressed key frame, and a maximum key frame size associated with the compression parameter value can be used to estimate spatial complexity (e.g., a degree of detail within the video). The number of predicted frames corresponding to the key frame can be used to estimate temporal complexity (e.g., an amount of motion within the video). A video complexity score can be calculated from these spatial and temporal complexity estimations and used to select a bitrate for subsequent video transmissions, enabling improved bandwidth usage and storage planning while maintaining video quality.

Abstract: Shared translation page tables are used for ownership transfer between software components having different execution privilege levels in a layered design. Each page table entry in a page table for the last translation level can include an ownership indicator to indicate the ownership of a physical page being mapped by the given page table entry by one or more software components. A single page table walk of the shared translation page tables can be performed to change the ownership of the physical page from one software component to another software component by updating the ownership indicator in a page table entry corresponding to that physical page.

Abstract: A hypervisor comprising a plurality of emulation routines is launched at a host. Individual emulation routines are used to perform respective tasks by a hypervisor on behalf of guest virtual machines of the host. A guest virtual machine initiates an operation which requires access to a virtual address. Subsequent to a determination by a memory management unit of the host that a virtual-to-physical address translation mapping for the virtual address which permits the operation initiated by the guest virtual machine is not available, a task is performed using a particular emulation routine, without initiating an analysis using the virtual address at the hypervisor to select the particular emulation routine from the plurality of emulation routines.

Abstract: Systems and methods are provided for generating a first model of a network based on a definition of the network, wherein the definition is used to deploy the network; generating a second model of the network based on a current deployment of the network; comparing the first model of the network and the second model of the network; and based on results of comparing the first model of the network and the second model of the network, generating output representing a discrepancy between the definition of the network and the current deployment of the network.

Abstract: This disclosure proposes systems and methods for processing natural language inputs using data associated with multiple language recognition contexts (LRC). A system using multiple LRCs can receive input data from a device, identify a first identifier associated with the device, and further identify second identifiers associated with the first identifier and representing candidate users of the device. The system can access language processing data used for natural language processing for the LRCs corresponding to each of the first and second identifiers, and process the input data using the language processing data at one or more stages of automatic speech recognition, natural language understanding, entity resolution, and/or command execution. User recognition can reduce the number of candidate users, and thus the amount of data used to process the input data. Dynamic arbitration can select from between competing hypotheses representing the first identifier and a second identifier, respectively.