Abstract: A file system manager implemented at a provider network identifies a storage device of a first group of storage devices of a provider network as an initial location of a file system object. Based on an access metric associated with the object, the file system manager initiates a transfer of contents of the object to a second storage device of a different storage device group, without receiving a client request specifying the transfer. In response to an access request received via a file system programmatic interface, contents of the object are provided from the second storage device. Based on a second access metric, the object is transferred back to the first group of storage devices.

Abstract: In response to determining that continuous data protection is to be enabled for a particular table of a database service, a service component verifies that automated transmission of change records of the table to a log-structured journal has been configured. A given change record comprises a before-image and an after-image associated with a committed database write, and is assigned a unique sequence number. In response to a determination to restore the table as of a specified point in time, a restore record set is identified from the journal with respect to a selected snapshot of the table. The restore record set includes change records which are not represented in the snapshot and are to be represented in the restored table. A restore result table is created using the selected snapshot and the restore record set.

Abstract: Distributed execution of transactions may be performed across shards of a scalable database table. Instructions to perform as part of a transaction with respect to one or more database tables may be received. The instructions may be evaluated with respect to metadata obtained for the database tables to determine an assignment distribution of computing resources of the database system to data that can satisfy the access request. The commit protocol for the transaction may be determined according to the assignment distribution of the computing resources. After a request to commit the transaction is received, the transaction may be committed according to the determined commit protocol.

Abstract: A stylus provides input via a touchscreen comprising a touch sensor and a display. Latency between placement of a stylus tip and corresponding presentation of visual indicia on the display is reduced or eliminated by determining a predicted path of the stylus tip during a stroke. Visual indicia is presented on the display, based on the predicted path. Inputs from the touch sensor may include hover events associated with detection of the tip while not in contact with the touchscreen and touch events associated with presence of the tip on the touchscreen. A machine learning network may be trained to determine the predicted path. A portion of the network may be trained to accept dynamic-length sequences of events and generate fixed length sequences, reducing subsequent network complexity. A hand may be detected and used to determine the predicted path. The end of a stroke may be predicted, reducing overshoot.

Abstract: Techniques for optimizing and deploying deep neural network (CNN) machine learning models for inference using static analysis are described. A method includes obtaining a deep neural network (DNN) machine learning (ML) model, generating an intermediate representation for the ML model, the intermediate representation including one or more nodes corresponding to one or more operators utilized by the ML model, identifying, for at least one node of the intermediate representation, an optimized schedule for at least one operator corresponding to the at least one node using a static analysis that is based on a hardware-specific cost model, generating an optimized intermediate representation using the optimized schedule that is optimized for execution on a hardware platform, and generating code corresponding to the ML model based at least in part on the optimized intermediate representation, wherein the code is specific to the hardware platform.

Abstract: Devices and techniques are generally described for security threat mitigation for generative machine learning models. In some examples, first prompt data including first data associated with a first natural language input and a first span may be determined. An LLM may determine first plan data using the first prompt data. The first plan data may include a call to the first API. A first classifier model may determine a first trust score for the first span. A first attention score may be determined for the first span and the first action plan. Second plan data may be generated based on at least one of the first trust score and the first attention score or the second trust score and the second attention score.

Abstract: Techniques implemented by a launch wizard of a cloud computing system to enable developers to create new workloads that can be launched by the launch wizard in the cloud computing system. Launch wizards offer users with guided experiences for identifying and deploying resources needed to support workloads of applications running in cloud computing systems. However, it is difficult for developers to create new workloads for the launch wizard to support. This disclosure describes a development platform through which developers create new workloads for the launch wizard to launch on behalf of users. The development platform provides a console through which developers define new workloads using a metadata language, and dynamically generates and presents a provisioning form that describes the new workloads. Once developers have finished defining new workloads, the new workloads are registered with the launch wizard such that users can use the launch wizard to launch the new workloads.

Abstract: Techniques for generating opinion-based content responsive to a user input are described. The system may receive a user input, and determine dialog context data corresponding to a dialog between a user and the system, and including the user input. The system may determine generation of content responsive to the user input requires opinion-based knowledge, and may extract entities from the dialog context data, and determine natural language data of a knowledge base that includes entities similar to the extracted entities. The system may processes the natural language data and the dialog context data to determine a subset of the natural language data that is responsive to the user input. The system may generate output data responsive to the user input using the responsive natural language data and the dialog context.

Abstract: A provider network implements a machine learning deployment service for generating and deploying packages to implement machine learning at connected devices. The service may receive from a client an indication of an inference application, a machine learning framework to be used by the inference application, a machine learning model to be used by the inference application, and an edge device to run the inference application. The service may then generate a package based on the inference application, the machine learning framework, the machine learning model, and a hardware platform of the edge device. To generate the package, the service may optimize the model based on the hardware platform of the edge device and/or the machine learning framework. The service may then deploy the package to the edge device. The edge device then installs the inference application and performs actions based on inference data generated by the machine learning model.

Abstract: This disclosure describes, in part, techniques for uploading network applications. For instance, system(s) may initially receive first data representing a first version of a network application and a first manifest that includes information describing the file contents of the first version. The system(s) may then store the first data and the first manifest within one or more databases. After the network application is updated to a second version, the system(s) may receive a second manifest that includes information describing the file contents of the second version of the network application. The system(s) may then use the first manifest and/or the second manifest to identify files that have been modified to create the second version of the network application. The system(s) may then receive second data representing the identified files and then reconstruct the second version of the network application using the first data and the second data.

Abstract: Techniques for adaptive machine learning training via in-flight feature modification are described. A training monitor captures training data during the training of a machine learning model, and a metric generator creates metrics such as feature importance metrics based on the data. A rule evaluation engine determines whether any modification conditions are met for any of the features based on the metrics, and based on such a determination can cause the in-flight training job to be modified.

Abstract: Techniques for utilizing dual-stack network addressing for compute instances hosted in an edge location of a cloud provider network along with communications service provider (CSP) network addresses are described. A first network address is assigned to the compute instance from a pool of network addresses of the cloud provider network, and a second network address is associated with the compute instance that is provided by the CSP network. A gateway of the edge location is updated to direct packets addressed to the second network address to the compute instance via use of the first network address.

Abstract: A network function execution intermediary obtains a result of execution of a set of network functions on a particular packet directed to a service from a client program running at a first host. Using the result, the packet is delivered to a request handler of the service, running at a second host; the path used for delivering the particular packet includes the network function execution intermediary. A networking manager of the first host obtains a rule for determining results of the set of network functions with respect to additional packets directed from the client program to the service. The networking manager causes, using the rule, another packet originating at the client program to be delivered to the request handler via a different path which does not include the network function execution intermediary.

Abstract: An insertion marker mechanism may be used to generate and evaluate insertion markers in media content. The system may place candidate markers at silent pauses and/or anomalies such as changes in pitch/timbre of an audio signal, and/or at points where a video has a blank screen. In response to a request for the media content, the system may place secondary content at a subset of the candidate markers. The system may select different insertion markers in response to different playback requests of the content. Based on feedback such as user engagement with the secondary content and/or drop rate, the system may determine which insertion markers correlate with better engagement and/or retention. Based on the feedback, the system may prioritize content insertion at better performing markers, while continuing to insert a smaller proportion of secondary content randomly at lower performing markers to monitor for changes in user behavior over time.

Abstract: A system comprising one or more computing devices implements a vehicle information extraction service. The vehicle information extraction service enables customers to maintain a model of a fleet of vehicles and collect vehicle information from heterogenous vehicles included in the fleet, wherein the vehicles are configured with different in-vehicle communication configurations. The vehicle information extraction service automatically generates scheme packets for data collection to be sent to the heterogenous vehicles, wherein are formatted in accordance with the different respective in-vehicle communication configurations of the heterogenous vehicles, without the customer of the vehicle information extraction service having to account for these differences.

Abstract: Various mounting systems for mounting heat sink apparatus to bare die processors are disclosed. The mounting systems include an upper plate, which may include a heat transfer portion, positioned in proximity to the upper surface of bare die processors to provide heat conduction away from the processors. The disclosed mounting systems secure the upper plate to the processors with balanced and centralized forces to inhibit tilting of the upper plate and reduce the risk of damaging the processors.