Microsoft Patente

An isolated platform for responsible Al is described. In various examples, a method is performed by a computing device. Output is generated by executing at least part of an application in an isolated virtual machine with an input-output virtualized accelerator, where the application is an artificial intelligence application. The isolated virtual machine is used to check the output. In response to the check being successful, the output is returned. In response to the check being unsuccessful, an error message is returned.

Methods, devices, systems, and computer program products are provided for adaptive determination of stylus tilt by selectively using gyroscopic and sensor data for accuracy of tilt determinations as signal conditions change. A device comprises an orientation detector that determines stylus tilt using a first tilt detector based on a first set of conditions detected at a first time. The device detects a second set of conditions at a second time. The device adapts tilt determination from the first tilt detector to a second tilt detector for determination of tilt at the second time. The device determines tilt at the second time using the second tilt detector. Adaptive tilt detection can vary tilt calculations by using different equations, using the same equation with different weights, blending the results of calculations using multiple equations, etc., to maintain the accuracy of tilt determinations as signal conditions change.

Methods, systems, and computer program products are provided that implement vector-related requests using query operators. For example, a system includes a database, a parser, a converter, an optimizer, and an execution engine. The database is associated with query operators that are non-vector-specific. The database stores a table with vector embeddings. The parser is configured to parse a request indicating a vector operation associated with the vector embeddings. The converter is configured to convert the vector operation into a logical operator tree comprising a representation of the request as a logical flow of the query operators, enabling the vector operation without vector-specific executable code or operators. The optimizer is configured to convert the logical operator tree into an executable plan. The execution engine is configured to execute the executable plan against the table with vector embeddings.

In various examples, an image search is performed based on a query image where the result includes images with semantic information that matches the query image. For example, image encoders are used to extract information from images that are stored in an index. Continuing this example, when a query is obtained the index is searched to determine potential matches, and a comparison is performed between the query image and the potential matches to determine images with semantic information matching the query image.

Examples solutions for evaluating network connectivity between computing resources within a computing environment include: generating a network packet for a proxy ping at a first host server that provides a first virtual machine (VM), the network packet being originally generated by the first host server and not generated by the first VM, the network packet being created to include a proxy ping flag within a header of the network packet; capturing a sending timestamp; receiving a response network packet generated by a second host server in response to interception of the network packet without the network packet being sent through to the second VM, the network packet having been identified for interception by the second host server based on identifying the proxy ping flag within the header of the network packet; capturing a response timestamp; and computing a latency value based on a difference between the sending and response timestamps.

Methods, systems, apparatuses, and computer program products are provided for protecting data in a memory of an integrated circuit (IC). A process token is obtained in a special purpose IC from a host that is external to and communicatively connected to the special purpose IC. The process token is stored in a first memory portion of the special purpose IC. In response to receiving a processing request from the host, the processing request is processed, and data generated by processing the processing request is written in a second memory portion of the special purpose IC. When a read request is received to read the data in the second memory portion, a determination is made whether the read request includes a read token that matches the previously stored process token. If the read token matches the process token, the data in the second memory portion may be returned to the host.

This disclosure generally relates to securely launching a hypervisor and subsequently validating that the hypervisor was securely launched. As is described herein, once a hypervisor has been initialized or has otherwise launched, a verification operation is performed. The verification operation may be used to ensure that the hypervisor was securely launched. When it is determined that the hypervisor was securely launched, one or more platform details are obtained. These platform details may then be stored in a memory device.

Techniques are described herein that are capable of performing role-based approval of a resource lifecycle event with regard to a cloud-based resource. A request to perform a resource lifecycle event with regard to a cloud-based resource is received. A designated approver is identified. The designated approver has authority to grant approval to perform the resource lifecycle event with regard to the cloud-based resource as a result of the designated approver having an approver role, which is authorized to grant the approval. An explicit approval is obtained from the designated approver. The explicit approval grants approval to perform the resource lifecycle event with regard to the cloud-based resource. As a result of an approval criterion being satisfied, performance of the resource lifecycle event with regard to the cloud-based resource is triggered. The approval criterion includes obtainment of the explicit approval.

Systems, methods, and computer program products are disclosed for thermal and power aware virtual machine (VM) placement in a data center including a set of rows separated by a set of aisles. When deploying a set of VMs in a data center for workload execution, a subset of candidate aisles that have sufficient estimated airflow capacity to host the set of VMs and a subset of candidate rows that have sufficient estimated power capacity to host the VMs are determined. A set of servers is selected in the subset of candidate aisles and in the subset of candidate rows based at least on a workload type associated with the workload, a distribution of workloads of the workload type in the subset of candidate aisles, and a distribution of workloads of the workload type in the subset of candidate rows. The set of VMs are deployed to the set of servers.

An anomaly detection model is tuned by training an anomaly detection condition tuning model using features of training input data provided to the anomaly detection model, training anomaly detection conditions used by the anomaly detection model to identify a detected anomaly in the training input data, and training data feedback identifying the detected anomaly as true or false. A first tuned training data condition is extracted for each feature by executing the anomaly detection condition tuning model on the features. A condition regression model is trained using second training data including the features of the training input data and each first tuned training data condition. Second tuned anomaly detection conditions are generated by executing the condition regression model on the new input data. The anomaly detection model is executable on the new input data using the second tuned anomaly detection conditions to identify an anomaly in the new input data.

Described are examples for detecting emerging patterns in data. A detection system for detecting patterns outside of supervised machine learning models is provided for determining similarity scores between transactions to detect the emerging patterns. Transactions in the pattern can be reviewed to determine whether to render decisions on the transactions or similar subsequently occurring transactions. A self-correcting detection system is also provided for using machine learning models to correct for emerging patterns in the transaction data.

The present disclosure generally relates to systems and methods for jointly predicting data table transformations and joins to prepare input data tables for various intelligence tools and applications. For example, the systems and methods discussed herein reason across transformations and joins within a set of data tables to increase the accuracy of both transformation and join predictions. The systems and methods discussed herein can jointly predict transformations and joins for a set of data tables by generating a graph representation that accounts for possible transformations of each data table in the set as well as join possibilities between the possible transformations. By solving this graph, the systems and methods discussed herein generate a sequence of transformations and joins for the set of data tables that jointly maximizes both the transformation probabilities and the join probabilities for the data tables in the input set of data tables.

Described are examples for rendering decisions based on machine learning (ML) model output. A set of segments for a historical set of data for a division of interest, and associated budgets for the decision of interest, can be obtained. For each segment in the set, a budget for incorrect decisions rendered based on output from the ML model can be computed. For each data entry in a current set of data, a current decision can be rendered based on a configured cutoff value and also a shadow decision based on the candidate cutoff value for a segment of the set of segments associated with the data entry. The candidate cutoff value can be promoted to replace the configured cutoff value, for rendering subsequent decisions for a subsequent set of data, based on comparing the shadow decisions for the current set of data based on the budget for incorrect decisions.

Improved hardware design testing techniques qualify a current design by integrating select tests completed on prior design iterations. A product design may be divided into design unit subcomponents, and multiple test benches may individually be configured to test different groups of subcomponents. The test benches may be run on a series of iterations of an evolving product design. Qualification of a most recent current iteration of the product design can be simplified by incorporating some test results from tests performed on prior iteration checkpoints of the design into the qualification of the current design by first combining test results across time for each test bench, and then combining test results across the set of test benches.

Embodiments relate to a computer-implemented method for managing zero-trust network communications. The method includes validating the identity of a first remote endpoint using a cryptographic credential received from the first endpoint and determining the first endpoint's authorization to communicate with a second remote endpoint. Upon successful validation and authorization, the method initiates the establishment of a network packet route between the first and second remote endpoints. Later, the method further initiates the destruction of the network packet route between the two endpoints.

Methods, apparatuses, and products for semantic search for retrieval-augmented generation, including: generating a semantic index for a table in response to a command to generate the semantic index for the table, including: generating a vector embedding of data in a corresponding first column; generating a vector index based on the vector embedding of the data in the corresponding first column; and generating a full-text index of a corresponding second column, wherein the semantic index comprises the vector index of each column of the one or more first columns and the full-text index of each column of the one or more second columns; and performing a semantic search of at least one column of the table indexed by the semantic index by applying a search string to the semantic index, wherein the search string corresponds to a query to a language model (LM).

Systems and methods for providing a drop tray engine for creating user-based drag-and-drop experiences are provided herein. In an aspect, a drop tray engine detects a drag-and-drop action performed on a content on a user interface of a client device. Responsive to detecting the drag-and-drop action, the drop tray engine determines one or more attributes of the content and determines predicted actions that a respective user is likely to perform with the content based on the one or more attributes. The drop tray engine then generates a drop tray containing action targets representing the predicted actions and displays the drop tray responsive to the drag-and-drop action on the user interface of the client device. Responsive to selection of an action target within the drop tray, the drop tray engine prompts a respective application to execute the associated action.

There is described herein a method for estimating splice loss in optical fibres, comprising the steps of: acquiring one or more side-view images of a splice between two fibres; processing the acquired images to extract features indicative of splice quality; extracting features from the processed images that are indicative of potential splice loss; inputting the extracted features into a machine learning (ML) model trained to estimate splice loss; generating an estimated splice loss value from the ML model based on the input features; and outputting the estimated splice loss.

The disclosed techniques provide a dynamic microphone pointer interface for generating contextual voice commands. A system provides an input that combines a voice input and a physical input gesture. In some embodiments, a system displays a UI with a movable microphone icon. The microphone icon can be moved by the physical input gesture to a region associated with a specific data object, such as a meeting, email, or file. While the microphone icon is indicating the selection of the region, the system can combine the voice input with that data object to complete tasks. This combination provides a technical improvement over a system that only receives a user's voice command by deriving context from the physical input gesture and combining that context with the user's voice command. This provides accurate and efficient input method that mitigates the need for repeated attempts to have a computer execute desired functions.

Aspects of the disclosure include a procedure for operating a quantum circuit. Aspects include performing Hamiltonian block encoding with unitary evolution on a first multiplicand and a second multiplicand resulting in a first intermediate operator and a second intermediate operator, respectively. Aspects include performing first gates on the first intermediate operator and second gates on the second intermediate operator resulting in a modified first intermediate operator and a modified second intermediate operator, respectively. Aspects include performing a commutator product formula to combine the modified first intermediate operator and the modified second intermediate operator resulting in a combination operator, and performing another first gates and another second gates on the combination operator resulting in a modified combination operator. The another second gates include a conjugate transpose of the another first gates. Two ancilla qubits are utilized on the quantum circuit.

An apparatus and method for adaptively calibrating a splicer are described. The method comprises receiving a first image of a portion of a hollow core fiber, the hollow core fiber comprising a splice produced by the splicer, wherein the portion of the hollow core fibre in the first image is proximate to the splice. The first image is analysed to detect a feature associated with the first image, the feature representing a property of the portion of the hollow core fibre. Using the feature, an adjustment to be made to a parameter of the splicer is computed, wherein the adjustment is computed so as to target one or more of a reduced optical propagation loss and a higher strength, of a subsequent splice of hollow core fiber which is produced by the splicer after initiating the adjustment. The method further comprises initiating the adjustment of the parameter of the splicer.

Patch updates enable intra-patch reversion from an updated binary to an initial binary in the event of a catastrophic runtime error. This is advantageous in so far as it allows the OS to revert back to a stable configuration with respect to a problematic binary without disabling non-problematic binaries included in the patch update. This in-turn provides an efficient alternative to the conventional approach of wholesale reverting from a patch update to an initial patch update during fault recovery following a catastrophic runtime error (e.g., failure to boot, etc.).

The techniques presented herein provide an interactive user interface for user-directed organization of content captures using visual indicators (e.g., tags) and a native content capture repository for systemwide interoperability. In various examples, a user can perform a gesture or other input to activate a tagging mode within a desktop environment. The user can then select a visual indicator (e.g., an emoji, text) from the tagging panel to attach to a content capture of the desktop environment. In one example, the visual indicator is generally associated with the content capture as a whole. In another example, the user utilizes an augmented cursor to place the selected visual indicator at a specific location in association with a specific object of visual content (e.g., an image, a block of text). Moreover, as content captures are stored natively, e.g., at the operating system level, external applications can access content captures via an application programming interface.

A system for providing representations of scene content is configurable to: (i) obtain one or more images including a first set of pixels and a second set of pixels; (ii) generate a first tone mapping operator using a first set of pixel values associated with the first set of pixels; (iii) generate a second tone mapping operator using at least a second set of pixel values associated with the second set of pixels; (iv) generate a first set of tone-mapped pixel values by applying the first tone mapping operator to the first set of pixel values; (v) generate a second set of tone-mapped pixel values by applying the second tone mapping operator to the second set of pixel values; (vi) generate an output image using the first and second sets of tone-mapped pixel values; and (vii) present the output image on the one or more displays.

Examples are disclosed that relate to controlling a device to enhance a digital image by selectively using different digital image enhancement algorithms based on adaptive device operation criteria that improves overall performance of the device. In one example, a digital image (204) is received from a camera (202). Sensor data (210) indicating a set of sensor parameter values (212) is received from a sensor subsystem (206). The digital image (204) is processed with a hardware-based image enhancement algorithm (222) to generate a first processed digital image (204) having higher image quality than the digital image (204) based on the set of sensor parameter values (212) not meeting adaptive device operation criteria (236). The digital image (204) is processed with an artificial intelligence (AI)-based image enhancement algorithm (226) to generate a second processed digital image (228) having higher image quality than the first processed digital image (224) based on the set of sensor parameter values (212) meeting the adaptive device operation criteria (236).

Various strategies for training a machine learning model are provided. One method computes a gradient matrix of a training loss function and applies a whitening transformation. Gradient normalization and whitening are employed to update weight vectors solely based on current gradient information without the need for exponential moving averages. A second method determines a state matrix to derive a truncated basis matrix composed of selected principal eigenvectors and additional orthogonal vectors and calculates an updated weight by projecting the gradient onto this low-dimensional subspace.

Improved techniques for remote access to a server's sensitive hardware debugging resources include a two-stage authentication of a remote debugger. The server's management controller may authenticate certificate data from the remote debugger and in response may bridge communications between the remote debugger and a debug port of the server's processor module. A security processor in the processor module may further authenticate the remote debugger by also authenticating certificate data provided by the remote debugger. After determining the remote debugger to be trustworthy, the security processor may enable the remote debugger to access hardware debugging resources within the hardware processor module.

The description relates to enhancing network security. One example can translate packet-level IDS rulesets into flow-level rulesets and can perform rule checking of flow summaries utilizing the flow-level rulesets.

A first AI agent is configured to obtain grounding information (e.g., information that will be used, by the first AI agent, to formulate a response to a query) from a first set of files and/or embeddings in a container associated with the AI agent. A second AI agent is configured to obtain grounding information from a second set of files and/or embeddings in a second container. Neither AI agent has access to files that are outside of the container associated with the AI agent. The container may be a folder into which a user can drag and drop files, and therefore the user may have visibility into (and control over) the information that is used as grounding information for each of the AI agents. An AI agent can be nested within a container for another AI agent to provide additional grounding information.

Devices, methods, and systems for determining localization lengths in hybrid superconductor-semiconductor devices using multiple fishbone structures having hybrid wires are described. An example method includes, using a measurement system coupled to a first fishbone structure, obtaining a first set of nonlocal conductance values associated with each of a first set of hybrid wires by selectively supplying voltages to a first set of gates associated with the first fishbone structure. The method further includes using the measurement system coupled to a second fishbone structure, obtaining a second set of nonlocal conductance values associated with each of the second set of hybrid wires by selectively supplying voltages to a second set of gates associated with a second fishbone structure. The method further includes determining the localization length associated with the hybrid superconductor-semiconductor device based on both the first set of nonlocal conductance values and the second set of nonlocal conductance values.

A system receives a user a configuration file including chatbot framework information for configuring a framework for a chatbot for a team of end users, uses the configuration file to determining a plurality of data sources relevant to the team; downloads document chunks from the data sources, processes the document chunks to generate metadata tags, and vectorizes the metadata tags to generate metadata embeddings for the plurality of document chunks. In response to receiving a user query, the system uses the metadata embeddings to select a collection of the document chunks and passes the collection of document chunks to a language model (LM) with the user query.

Methods, apparatuses, and products for enhancing retrieval augmented generation with subject matter expertise, including: generating, based on a semantic persona, documented learnings associated with content ingested by a reflection engine; identifying, from a knowledge base that is used for retrieval augmented generation, one or more entries within the knowledge base that are associated with the documented learnings; and generating a mapping between the documented learnings and the one or more entries, wherein retrieval augmented generation ('RAG') is performed by a large language model ('LLM') using, based on the generated mappings, the knowledge base and the documented learnings.

Systems, methods, devices, and computer readable storage media described herein provide techniques for automatic compaction of data for table formats. In an aspect, a statistic is determined for a plurality of candidates. A trait is determined based on the statistic. The plurality of candidates are ranked with respect to a compaction objective based on the trait. Performance of a compaction action is caused with respect to a candidate based on a ranking of the candidate.

Methods, systems, and computer storage media for providing anomaly detection using an anomaly detection engine of a security management system are described. The anomaly detection engine supports anomaly detection that refers to the process of identifying command-line activities that deviate from established patterns of normal behavior within a large-scale environment, such as data centers. The anomaly detection engine operates based on an advanced framework specifically designed to enhance the security of large-scale environments by detecting anomalous and potentially malicious command-line activities. By integrating global and local context analysis, employing efficient data filtering and distributed computing, and leveraging advanced models for pattern recognition, the anomaly detection engine provides a scalable approach to anomaly detection. Operationally, the anomaly detection engine can be implemented as based on a dual-layer detection architecture, use of statistical models in command-line analysis, granular and context n-gram tokenization, adaptable anomaly scoring and thresholding, scalability, and continuous compliance monitoring.

Examples of the present disclosure describe systems and methods for transmit-side scaling. In an example, transmit-side configuration information is received by a host operating system from a client operating system, where the transmit-side scaling configuration information specifies a manner in which data packets transmitted by the host operating system are to be distributed to a network interface card of the host operating system. The transmitting-side configuration scaling information is implemented in an outbound transport table of the host operating system. When a data packet is received by the host operating system, the host operating system evaluates the data packet using the outbound transport table. Based on the evaluation, the data packet is sent using a specified transmission queue of the network interface card.

Methods and apparatuses for generating mixed-modality summaries of mixed-modality data subject to constraints that vary over time, end users, output device types, and operating environments are described. A mixed-modality summary generation system generates mixed-modality embeddings within a joint embedding space using the mixed-modality data, determines user-derived constraints and output device constraints, determines a coreset of the mixed-modality embeddings within the joint embedding space based on the user-derived constraints and output device constraints, generates a mixed-modality summary using the coreset, and outputs the mixed-modality summary using an output device. Based on the user-derived constraints and the output device constraints, the mixed-modality summary generation system may identify joint-modality or single-modality embeddings, wherein each embedding comprises a joint-modality or single-modality embedding within a threshold distance to one of the embeddings within the coreset of the mixed-modality embeddings.

Techniques are directed to application authentication based on registration with an application registration service. In an example, an application frontend registers with the application registration service and receives a registration key. The application frontend provides a proof of possessing the registration key and a credential of application frontend or an associated account to an authentication service. The authentication service determines an authenticity of the application frontend or the account based on the credential and determines a validity of the proof of possession. Responsive to the authentication and validation, the authentication service provides an artifact to application frontend. The application frontend utilizes the artifact to access an application backend and/or a resource thereof. The artifact is bound to a particular instance of the application frontend in a manner that prevents other instances of the application frontend not possessing the registration key from utilizing the artifact to access the application backend.

A system having scalable sockets to support User Datagram Protocol (UDP) connections identifies a plurality of UDP connections, wherein a plurality of remote clients connect to corresponding ones of the plurality of UDP connections. Each one of a plurality of UDP sockets is associated with a corresponding one of the plurality of UDP connections. A network stack lookup for UDP packets in network traffic is performed using a network stack to identify the UDP socket corresponding to the remote client associated with each of the UDP packet. The UDP packets are buffered with a send buffer and a receive buffer for the UDP socket corresponding to the remote client associated with the UDP packets as determined by the network stack lookup to support communication over the plurality of UDP connections using the plurality of UDP sockets. The system thereby operates more efficiently and/or is more scalable.

A system implements techniques for using traffic volume associated with a service in order to proactively manage the capacity of a resource that supports the service. More specifically, the traffic volume is used as an indicator to determine, or project, a change in capacity that may be needed to avoid the under-provisioning of the resource. Alternatively, the traffic volume is used as an indicator to determine, or project, a change in capacity that may be preferred by a tenant that operates the service to avoid the over-provisioning of the resource.

Technologies are disclosed for providing visual feedback during touch-based operations on user interface ("UI") elements. Through implementations of the disclosed technologies, visual feedback can be provided to users during certain touch-based operations on UI elements presented by touch-enabled computing devices. The visual feedback can provide confirmation to users that certain touch-based operations on UI elements were performed successfully. The visual feedback provided by the disclosed technologies can make it easier for users to select a UI element, to understand that actions taken to select a UI element were successful, and to successfully perform operations on UI elements once they have been selected. These benefits can reduce the number of times users need to attempt certain touch-based operations on UI elements. This, in turn, can reduce the utilization of computing resources, such as memory and processor cycles, by touch-enabled computing devices implementing the disclosed technologies.