Abstract: Techniques for hosting adding and warming a host are described. In some instances, a method of determining that at least one group of hosts is to be increased by adding an additional host to the group of hosts; sending a request to the group of hosts for a list of machine learning models loaded per host of the group of hosts; receiving, from each host, the list of loaded machine learning models; loading at least a proper subset of list of loaded machine learning models into random access memory of the at least one group; receiving a request to perform an inference; routing the request to the additional host of the group of hosts; performing an inference using the additional host of the group of hosts; and providing a result of the inference to an external entity is described.

Abstract: Techniques for intelligent compute resource selection and utilization for machine learning training jobs are described. At least a portion of a machine learning (ML) training job is executed a plurality of times using a plurality of different resource configurations, where each of the plurality of resource configurations includes at least a different type or amount of compute instances. A performance metric is measured for each of the plurality of the executions, and can be used along with a desired performance characteristic to generate a recommended resource configuration for the ML training job. The ML training job is executed using the recommended resource configuration.

Abstract: Disclosed are various embodiments for distributing data items within a plurality of nodes. A data item that is subject to a data item update request is updated from a master node to a plurality of slave notes. The update of the data item is determined to be locality-based durable based at least in part on acknowledgements received from the slave nodes. Upon detection that the master node has failed, a new master candidate is determined via an election among the plurality of slave nodes.

Abstract: A request associated with reviewing content for a field of interest is received. A confidence is determined associated with the content including the field of interest. A machine learning (ML) model determines a first confidence associated with the content includes the field of interest. The field of interest is transmitted for review in instances where the first confidence is less than a confidence threshold. After review, an indication associated with a reviewer reviewing the content and the first confidence associated with the ML model identifying the field of interest is updated to a second confidence.

Abstract: Querying databases may be performed with references to machine learning models. A database query may be received that references a machine learning model and database. In response to the query, the machine learning model may provide information which may be returned as part of a result of the query or may be used to generate a result of the query. The machine learning model may be generated in response to a request to generate a machine learning model that includes a database query that identifies the data upon which a machine learning technique may be applied to generate the machine learning model.

Abstract: Data to be stored in a data block for a columnar database table may be compressed according to a multi-level compression scheme. Data to be stored in the data block may be received. The data may be compressed according a column-specific compression technique to produce compressed data. The compressed data may then be compressed according to a second compression technique different than the column-specific compression technique to produce multi-level compressed data. The multi-level compressed data may be stored in the data block. When reading from the data block, multi-level compressed data may be decompressed according to the column-specific compression technique and the default compression technique applied to the data.

Abstract: Techniques for making machine learning inference calls for database query processing are described. In some embodiments, a method of making machine learning inference calls for database query processing may include generating a first batch of machine learning requests based at least on a query to be performed on data stored in a database service, wherein the query identifies a machine learning service, sending the first batch of machine learning requests to an input buffer of an asynchronous request handler, the asynchronous request handler to generate a second batch of machine learning requests based on the first batch of machine learning requests, and obtaining a plurality of machine learning responses from an output buffer of the asynchronous request handler, the machine learning responses generated by the machine learning service using a machine learning model in response to receiving the second batch of machine learning requests.

Abstract: Methods, systems, and computer-readable media for automated problem detection for machine learning models are disclosed. A machine learning analysis system receives data associated with use of a machine learning model. The data was collected by a machine learning inference system and comprises input to the model or a plurality of inferences representing output of the machine learning model. The machine learning analysis system performs analysis of the data associated with the use of the machine learning model. The machine learning analysis system detects one or more problems associated with the use of the machine learning model based at least in part on the analysis. The machine learning analysis system initiates one or more remedial actions associated with the one or more problems associated with the use of the machine learning model.

Abstract: A request to perform a batch of operations is provided to a distributed database. The request comprises instructions for validating a condition. An association between the request and a unique identifier is stored. An item in the distributed database is locked and the condition is validated. The system that initiates processing of the batch of operations. A second request, comprising the identifier, is received. The second request is responded to by providing information indicative of the status of processing the first request, based on the stored association. The lock is released when processing of the first request has completed.

Abstract: Techniques for active learning-based data labeling are described. An active learning-based data labeling service enables a user to build and manage large, high accuracy datasets for use in various machine learning systems. Machine learning may be used to automate annotation and management of the datasets, increasing efficiency of labeling tasks and reducing the time required to perform labeling. Embodiments utilize active learning techniques to reduce the amount of a dataset that requires manual labeling. As subsets of the dataset are labeled, this label data is used to train a model which can then identify additional objects in the dataset without manual intervention. The label data can be added to an augmented manifest, the augmented manifest can be used to filter the dataset to perform further labeling jobs on the same or different subsets of the dataset.

Abstract: Disclosed are various embodiments for distributing data items. A plurality of nodes forms a distributed data store. A new master candidate is determined through an election among the plurality of nodes. Before performing a failover from a failed master to the new master candidate, a consensus is reached among a locality-based failover quorum of the nodes. The quorum excludes any of the nodes that are in a failover quorum ineligibility mode.

Abstract: Systems and techniques are disclosed for a centralized platform for enhanced automated machine learning using disparate datasets. An example method includes receiving user specification of one or more data sources to be integrated with the system, the data sources storing datasets to be utilized to train one or more machine learning models by the system, and the datasets reflecting user interaction data. A dataset is imported from the data source, and machine learning models are automatically trained based a particular machine learning model recipe of a plurality of machine learning model recipes. A first trained machine learning model is implemented, with the system being configured to respond to queries based on the implemented machine learning model, and with the responses including personalized recommendations.

Abstract: Techniques for hosting machine learning models are described. In some instances, a method of receiving a request to perform an inference using a particular machine learning model; determining a group of hosts to route the request to, the group of hosts to host a plurality of machine learning models including the particular machine learning model; determining a path to the determined group of hosts; determining a particular host of the group of hosts to perform an analysis of the request based on the determined path, the particular host having the particular machine learning model in memory; routing the request to the particular host of the group of hosts; performing inference on the request using the particular host; and providing a result of the inference to a requester is performed.

Abstract: A distributed database maintains a table on a first plurality of partitions. A request to restore the table to a point-in-time is received. The database determines, based on log data of the partitions, a maximum version number of an operation processed by the partitions. The log data is processed to exclude, from the restoration, operations whose transactions were started after the point-in-time, by setting the version number of those operations to be greater than the maximum version number. The log data is then applied to a second plurality of partitions, where the version number of each applied operation is less than or equal to the determined maximum version number.

Abstract: A distributed database system maintains data for a database client by storing data on a plurality of storage nodes. Upon receiving a request from the database client in a first format, the database system translates the request to a second format and sends the translated request to a storage subsystem. The storage subsystem generates an exception if the translated request cannot be successfully completed. The distributed database system resends the translated request on behalf of the database client if the exception corresponds to a request that can be retried, and continues to resend the translated request until a first of an expiration of a predetermined time period or until the request completes successfully. The distributed database system sends a response to the database client based on the resent database request.

Abstract: A dishwasher having a water pump, a water conduit connected to the water pump for distributing water in the dishwasher tub, a door for closing and opening the dishwasher, and a detergent dispenser located on the inside of the door. The detergent dispenser comprises a closeable cavity for receiving a detergent, an outlet into the tub, and a sealed dispenser water inlet configured to be detachably connected to the water conduit when the door is closed. Operation of the pump causes pressurized water to be directed directly to the inside of the detergent dispenser while the detergent dispenser is in a closed state.

Abstract: Fine-grained virtualization provisioning may be performed for in-place database scaling. Computing resource utilization for a database on a host system is obtained for a period of time. The computing resource utilization may be evaluated with respect to a target capacity for the database. If a scaling event is detected based on the evaluation, a modified target capacity may be determined and used to make an adjustment of the computing resources permitted to be used by the database.

Abstract: A computer system that provides clients access to pooled resources in order to provide computing or data storage services may receive service requests (which explicitly or implicitly include requests for pooled resources), may service at least some of those requests, and may determine pricing for the serviced requests. The pricing for each request may be dependent on whether it was serviced using a portion of a resource pool that was reserved for the use of the client on whose behalf it was received or using burst capacity (e.g., unreserved or otherwise idle capacity within the resource pool). Pricing for the use of reserved capacity may be fixed, regardless of the amount of reserved capacity used. Pricing for burst capacity may depend on actual use, and may be demand-based (e.g., using a spot-market-based dynamic pricing model). Clients with reserved capacity may optionally request access to burst capacity.

Abstract: Scalable architecture for propagating updates may be implemented for data replicated from a data set. A node may receive updates to items in a data set that have been committed to the data set. The node may determine whether the update should be applied to a replicated portion of the data set. For updates that should be applied, the node may identify another node that hosts the replicated portion of the data set and send a request to the other node to perform a conditional atomic operation to apply the update to the item in the replicated portion of the data set. The condition may compare a version identifier associated with an update and a current version identifier for the item at the other node. If the condition evaluates true, then the update to the item in the replicated portion may be performed.

Abstract: A distributed data store may implement passive distribution encryption keys to enable access to encrypted data stored in the distributed data store. Keys to encrypt a data volume stored in the distributed data store may be encrypted according to a distribution key and provided to a client of the distributed data store. Storage nodes that maintain portions of the data volume may receive the encrypted key from a client to enable access to the data volume. The storage nodes may decrypt the key according to the distribution key and enable access to the data volume at the storage nodes. In some embodiments, a key hierarchy may be implemented to encrypt the keys that provide access to the encrypted data. The key hierarchy may include a user key.