Abstract: Systems and methods are disclosed for identifying manipulated items using neural networks. Systems generate, using a first neural network, indications of an object of interest from the two or more objects in one or more images. Systems identify, using a second neural network, structured identifiers based, at least in part, the one or more images. Then, the system selects a structured identifier associated with the object of interest from the structured identifiers based, at least in part, on the indications to the object of interest.

Abstract: Techniques for emulating a configuration space may include emulating a set of configuration registers in an integrated circuit device for a set of functions corresponding to a type of peripheral device. The type of peripheral device represented by the integrated circuit device can be modified by changing the set of configuration registers being emulated in the integrated circuit device. Multiple sets of configuration registers can also be emulated to support different virtual machines or different operating systems.

Abstract: A peripheral device includes one or more processors and a memory storing program instructions that when executed implement an extension manager of a virtualized computing service. The extension manager establishes a secure network channel for communications between the peripheral device, which is located at a premise external to a provider network, and a data center of the provider network. The extension manager assigns a network address of the substrate network of the service to a hardware server at the external premise. The substrate address is also assigned to an extension traffic intermediary at the data center. In response to a command directed to the virtualized computing service, one or more compute instance configuration operations are performed at the hardware server.

Abstract: Dynamic translation of data records is performed. A client of a translation service may request translation of binary data according to a binary data protocol to be output according to an output schema different from the binary data protocol. A dynamic translation engine may be configured to connect to an input data stream implementing the binary data protocol and connect to an output data consumer or storage device. The dynamic translation engine may then identify records on the source data stream according to the binary protocol, extract keys and values for individual fields of records, then output key-value pairs or alternative binary encodings to output records formatted according to the output schema. The binary data protocol and output schema may be dynamically defined and optionally updated during translation.

Abstract: Disclosed are various embodiments for predicting excess capacity for radio access network (RAN) edge systems. In one embodiment, resource consumption data is generated by monitoring resource consumption of a plurality of implementations of a distributed unit (DU) and/or centralized unit (CU) used in RANs. A maximum level of resource consumption is predicted for a particular implementation of the DU/CU to be used in a particular RAN based at least in part on the resource consumption data. Excess resource capacity is predicted on a RAN-enabled edge server on which the particular implementation of the DU/CU is deployed for a cell site of the RAN based at least in part on the maximum level of resource consumption.

Abstract: Metadata for external table sources of a local materialized view is transactionally derived. An update to the metadata is identified. Data objects storing respective portions of external tables are determined according to the update to the metadata. The data objects are accessed to obtain data from the external tables in order to update the local materialized view.

Abstract: For a given cross-data-store transaction request at a storage service, a coordinator transmits respective voting transition requests to a plurality of log-based transaction managers (LTMs) configured for the respective data stores to which writes are directed in the transaction. The LTMs transmit responses to the coordinator based on data-store-specific conflict detection performed using contents of the voting transition requests and respective data-store-specific state transition logs. The coordinator determines a termination status of the cross-data-store transaction based on the LTMs' responses, and provides an indication of the termination status to the LTMs.

Abstract: Systems and methods for end of arm tools having belt-driven engagement members and related item manipulation devices are provided. In one embodiment, an example item manipulation device may include a housing, a suction cup assembly having a first suction cup, where the suction cup assembly is positioned at a distal end of the housing, a first engagement member having a first arm portion configured to rotate with respect to the housing, a first belt coupled to the first arm portion, and a first actuator configured to actuate the first belt to cause the first arm portion to rotate with respect to the housing.

Abstract: An elastic field programmable gate array (FPGA) system may include a plurality of FPGA devices connected in a physical ring topology. Each FPGA device may include a user programmable logic partition and a platform logic partition. The platform logic for each FPGA device can be selected by the client from a plurality of platform logic designs having progressively fewer components. Each platform logic design can include a platform management control block coupled to a microcontroller to implement device management functions and programming of the FPGA device. The size of the user programmable logic partition and the amount of available FPGA resources can vary for each FPGA device based on the platform logic design selected for that FPGA device.

Abstract: Techniques for generating and providing phrases built using multi-modal inputs are described herein. These techniques may include generating one or more phrase trees for use in parsing options for user selections through multi-modal inputs to iteratively build an input phrase such as an open-ended phrase. The multi-modal inputs may include gaze inputs, touch inputs, and other inputs to select terms for building a phrase. The selected terms may be used to build a phrase using trees of terms arranged in sequential nodes based on frequently used input phrases to a model.

Abstract: A large language model (LLM) for automated database query resolution is provided. The LLM is configured to receive natural language prompts and generate database queries or direct answers to the prompts. To assist the LLM in generating the response, contextual information (e.g., historical queries, table metadata, etc.) may be captured, converted into embeddings, and stored in a vector database. When a natural language prompt is received, the prompt is also converted into an embedding and a nearest neighbor search may be performed to identify the most relevant contextual information. This most relevant information is then provided to the LLM. Additionally, one or more knowledge graphs may also be generated, with nodes representing tables and indications of relationships between the nodes being provided. These knowledge graphs may also be provided to the LLM.

Abstract: Techniques for cross-regional resource linking in a federated data lake management and integration system. The techniques include receiving a query and identifying a referenced database object. First, metadata about this object is obtained from a local data lake metastore. If this metadata includes a cross-regional resource link, the techniques then retrieve second metadata from a remote data lake metastore. Based on this second metadata, a query execution plan is generated. This plan guides a distributed data processing system to execute data processing jobs on data stored in a distributed data storage system. Finally, the data responsive to the query is returned. The techniques efficiently integrate and manage data lake metadata resources across multiple regions.

Abstract: Systems, devices, and methods are provided for cryptographic control access and revocation of data. When data is to be shared between a data producer and a data consumer, a key vault is used to store a consumer-producer key pair that is used to generate a portion of key material that is used to derive an envelope key. Data producer is able to generate the envelope key for encryption using the consumer-producer public key, whereas the data consumer relies on the key vault to provide the portion of key material to produce the envelope key for decryption. Access to data can be revoked through the destruction or of the consumer-producer private key stored in the key vault.

Abstract: System and methods are described to connect an agent of one region to a contact request in another region via a proxy service. In some examples, a first contact service instance of a first region may obtain a contact request from a contact requestor, where the contact request includes a set of attributes. The first contact service instance selects an agent of a second contact service instance of a second region to process the contact request, whereby a proxy service of the first contact service instance sends a request to the second contact service instance to release the agent to process the request. The first contact service instance/proxy service may obtain an indication that the agent is approved to process the contact request. The first contact service instance may then connect the contact requestor and the agent to process the contact request.

Abstract: Systems and methods are disclosed for induction of items directly onto shuttles. An example system for item induction onto a shuttle may include a track, a shuttle having a conveyor and a permanent magnet, the shuttle configured to move along the track, where the conveyor is configured to support an item thereon, and an electromagnet disposed along the track, where the shuttle is propelled via interaction between the electromagnet and the permanent magnet. The system may include an induction station disposed adjacent to the track, the induction station having a first sensor configured to detect a machine-readable code on the item, a second sensor configured to determine that the item is present on the conveyor of the shuttle, and a display configured to present a user interface, where the item can be manually placed onto the conveyor of the shuttle by an operator at the induction station.

Abstract: Disclosed are various embodiments for extending cloud-based virtual private networks to user equipment on radio-based networks. In one embodiment, a request is received from a client device for service from a radio-based network. The client device is provided with access to a virtual private cloud network through the radio-based network in response to receiving the request from the client device for service from the radio-based network. Encapsulated network traffic is forwarded from the client device to the virtual private cloud network via a network link between the radio-based network and a cloud provider network that hosts one or more resources on the virtual private cloud network.

Abstract: An access management policy and security information indicative of one or more security constraints may be received, by a service, from an entity, such as a customer. The access management policy and the security information may be received in association with a request for the service to perform a policy validation check for validating the policy. The service may perform a permissions comparison that compares a permissiveness of the access management policy to the one or more security constraints to generate a permissions comparison result. The service may generate, based on the permissions comparison result, a binary policy validation check result that may be indicative of either passing or failing the policy validation check. When the access management policy fails the policy validation check, the service may determine a given allow statement within the access management policy that caused the access management policy to fail the policy validation check.

Abstract: Systems and methos are described relating to distributing agents to different queues provided by a contact service. The contact service may define a number of different queues for processing different customer requests, such as may be routed to different agents, with each queue associated with a priority value (higher priority get assigned agents first) and a visibility threshold (how long a request will sit before it is picked up by an agent). The described techniques add a time-to-service level (SL) goal, per queue to determine when queues are over and underperforming. In the case queues are overperforming, various techniques, such as control systems techniques, may be used to determine probabilities by which agents can skip being 10 assigned to the over performing queues, to be assigned to lower priority underperforming queues, to increase meeting the customer SL goals.