Abstract: A human-model collaborative annotation system for training human annotators includes a database that stores images previously annotated by an expert human annotator and/or a machine learning annotator, a display that displays images selected from the database, an annotation system that enables human annotators to annotate images presented on the display, and an annotation training system. The annotation training system selects an image sample from the database for annotation by a human annotator, receives one or more proposed annotations from the annotation system, compares the human annotator's one or more proposed annotations to previous annotations of the image sample by the expert human annotator or machine learning annotator, presents attention maps on the display to draw the human annotator's attention to any annotation errors identified by the comparing, and selects a next training image sample from the database based on any errors identified in the comparing step.

Abstract: A computing apparatus is provided comprising a client and a server. The computing apparatus is configured to: obtain a machine learning code; split the machine learning code into a first part and a second part; execute the first part of the machine learning code on the server; execute the second part of the machine learning code on the client; and output a result of the machine learning code. In this way, the machine learning code may be split and executed over both the client and the server in an efficient way.

Abstract: The application disclose an objective function solving method and apparatus, and a computing device cluster, and belong to the field of cloud computing technologies. The method includes: receiving a solving requirement input, where the solving requirement includes an objective function, a decision variable, and a constraint condition; determining, based on the solving requirement, a simplex method as a solving method, to solve the objective function; in a process of solving the objective function using the simplex method, after solving the objective function according to a first pricing strategy using the simplex method, determining, based on an objective improvement on the objective function by the current solving in the simplex method, a second pricing strategy for solving the objective function in a next iteration; and solving the objective function according to the second pricing strategy. According to this application, objective function solving efficiency can be improved.

Abstract: A data storage method is provided, and includes: obtaining a plurality of data fragments, where the plurality of data fragments belong to a first data segment (S201) in a first data stream; in response to a first storage parameter indicating that the first data segment (S201) meets a storage condition (202), storing the plurality of data fragments concurrently in a distributed storage system (200), where the first storage parameter is determined based on a storage case of a data segment in front of the first data segment (S201) in the first data stream, and the distributed storage system (200) includes a plurality of storage nodes (231, 232, 233, 234); and updating the first storage parameter (203) based on a storage case of the plurality of data fragments. In this way, a throughput of the distributed storage system (200) can be improved, and efficient data storage can be implemented.

Abstract: This application provide a resource scheduling method and a related device, and relate to the field of communication technologies. The method includes: building, by obtaining resource information of each node in a cluster and resource information of a container in each node, a resource profile of a microservice deployed on each node, where a container engine is installed on each node; and when determining that a first microservice deployed in the cluster is a microservice on which resource adjustment needs to be performed, generating, based on a resource profile of the first microservice, resource information of a first node on which the first microservice is deployed, and resource information of a container in the first node, a resource adjustment instruction associated with the first microservice, and adjusting, based on the container engine and the resource adjustment instruction, a resource of a pod associated with the first microservice.

Abstract: This application relates to an image retrieval method, a model training method, an apparatus, and a storage medium. The method may include: segmenting a target region in a to-be-retrieved image through a first neural network, to obtain at least one first feature, where the first feature is an intermediate feature extracted in a process of segmenting the target region; processing the to-be-retrieved image and the at least one first feature through a second neural network, to obtain a feature corresponding to the target region; generating, based on the feature corresponding to the target region, target code corresponding to the to-be-retrieved image; and retrieving a target image from a preset image set based on the target code. In this way, through information exchange between the first neural network and the second neural network, utilization of a feature extracted by the first neural network is improved, effectively improving accuracy of similarity retrieval.

Abstract: A database management system, comprising: a storage adapted for storing: a plurality of data objects organized according to a data model, and a plurality of stream operator wrappers each wrapping a stream operator and having at least one port for receiving, via a network, instructions for: scheduling activation of the wrapped stream operator, connecting the wrapped stream operator with another stream operator wrapped by another of the plurality of stream operator wrappers, and/or deploying the wrapped stream operator; a processing circuitry for executing the plurality of stream operator wrappers.

Abstract: A method for training a neural network, and a related device are provided. The method may be applied to a scenario in which a neural network is trained in the field of artificial intelligence. When an Nth round of training of the neural network is being performed, a first computational graph may be obtained; and after it is determined that a first compiled code corresponding to the first computational graph has been stored in a system, the first compiled code may be directly executed, where the first compiled code is generated during execution of an Mth round of training of the neural network, and M is less than N. Because there is no need to perform an operation of converting the first computational graph into an intermediate representation, and obtaining a compiled code based on the intermediate representation, overheads of computer resources are reduced.

Abstract: This application provides a resource scheduling method, including: when a first scheduling request message is obtained, determining a first resource server from a resource pool based on a first quantity of resources requested by the first scheduling request message, and scheduling the first quantity of resources in the first resource server, where the resource pool includes at least one resource server, and the first scheduling request message requests a resource for a first-type task; and when a second scheduling request message is obtained, schedule a resource for a task corresponding to the second scheduling request message, determining a second resource server from the resource pool based on a second quantity of resources requested by the second scheduling request message, and scheduling a third quantity of resources in the second resource server, wherein the second scheduling request message requests a resource for a second-type task.

Abstract: In an example application instance deployment method, a global management platform receives quality of service (QoS) requirement information from a first client. The QoS requirement information includes a first delay requirement, a second delay requirement, and a first quantity of connections. The second delay requirement is better than the first delay requirement, and the QoS requirement information is entered by a first user to the first client. The global management platform selects a first available site that meets the first delay requirement from managed sites. The global management platform deploys one or more first application instances on the first available site. A quantity of connections that can be established to the first application instance is less than the first quantity of connections.

Abstract: In a method for controlling a cloud desktop service application, and electronic device receives first clock-punching data sent by a terminal device. The first clock-punching data comprises a first clock-punching time and a clock-punching location of a user, and the clock-punching location is a geographical location of the terminal device at the first clock-punching time. The electronic device determines a start time of the cloud desktop service application based on the first clock-punching data and the first location, and start, based on the start time, the cloud desktop service application on a first desktop of the electronic device.

Abstract: This application discloses an object recognition method and apparatus, and a storage medium, and belongs to the field of computer vision. The method includes: obtaining to-be-processed visual data and indication information of at least one target object to be recognized; obtaining semantic information based on the indication information of the at least one target object, where the semantic information is semantics used to describe the at least one target object; and recognizing the target object in the visual data based on an object recognition model and the semantic information. In this application, flexibility of object recognition can be improved.

Abstract: This application provides a data sharing method, including: A first data management system receives a first query statement; obtains, based on the first query statement, a second query statement by using a software development kit of a second data management system; and then accesses a unified open access service of the second data management system based on the second query statement, to obtain proprietary-format data in the second data management system. In the method, a data management system may provide a software development kit SDK to the outside, such that another data management system such as a big data engine can access a unified open access service by invoking the SDK, so as to support sharing of proprietary-format data with the outside, breaking a limitation that proprietary-format data is shared only within a system, and improving usability.

Abstract: A microservice migration method is applied to a cloud platform configured to manage an instance that provides a microservice, and includes: obtaining stored first microservice configuration information of a target microservice from a first configuration center running on a first instance; generating second microservice configuration information of a migrated target microservice based on the first microservice configuration information; and sending the second microservice configuration information to a second configuration center running on a second instance, where the second configuration center is configured to store configuration information of the migrated target microservice. In this way, during microservice migration, microservice configuration information may be obtained from an original configuration center, and then based on the microservice configuration information, microservice configuration information adapted to a new configuration center is obtained.

Abstract: A data provider encrypts source data to obtain a ciphertext of the source data, and uploads the ciphertext of the source data to a data storage platform for storage. Subsequently, in response to a data application request of a data consumer, the data provider encrypts a storage address of the ciphertext of the source data by using a public key of the data consumer, and uploads the encrypted storage address to a blockchain network. The data consumer obtains the ciphertext of the source data from the data storage platform. In the process, a blockchain decentralization capability is used to implement secure and trusted data exchange by using a smart contract that is public and commonly visible to a plurality of parties.

Abstract: This application discloses a data management method, apparatus, and system, and a storage medium, and belongs to the communication field. The method includes: obtaining first description information, where the first description information is used to describe a first row group, the first row group is a row group stored by a first storage system within a specified time period, and the first row group includes at least one row of data; and storing the first row group in a second storage system based on the first description information. This application can reduce operation costs and maintenance difficulty.

Abstract: In a method for AI model optimization, an optimization device receives an original AI model and search configuration information that comprises a plurality of search items each indicating its search categories for performing optimization information search on the original AI model. The device obtains a plurality of search operators corresponding to the plurality of search items, and arranges the search operators in an operation sequence based on the search configuration information. The device then executes the search operators in the arranged operation sequence on the original AI model to obtain an optimized AI model. In the execution of the operation sequence, each search operator, except for the first search operator in the operation sequence, is executed utilizing operation results of a preceding search operator in the operation sequence, the operation results including generated network structures and search space information.

Abstract: This application relates to the field of cloud computing, and specifically, to a method for providing an edge service for a terminal by using a resource of an edge resource cluster in a cloud computing system. The cloud computing system includes a central resource cluster and at least one edge resource cluster. The method includes: a management node deployed in the central resource cluster determines a target execution node based on an edge service application range that is specified by a tenant or that is determined by the management node based on information about a tenant; the target execution node determines, according to an edge service policy sent by the management node, a target edge node from at least one edge node deployed in the at least one edge resource cluster; and the target execution node further forwards an edge service request to the target edge node.

Abstract: A model training system includes a cloud data storage platform and a cloud model training platform. The cloud data storage platform is configured to: store training data; and receive a training data calling request, and export training data corresponding to a data calling instruction to the cloud model training platform based on the training data calling request. The cloud model training platform is configured to: receive a model training creation instruction to obtain a to-be-trained model; generate the training data calling request, and send the training data calling request to the cloud data storage platform; and train the to-be-trained model by using the training data exported from the cloud data storage platform, to obtain a training result model. In the technical solutions of the present invention, a risk of leaking training data can be reduced.

Abstract: A border gateway receives a first overlay packet sent by a remote acceleration gateway, where a first service request packet is encapsulated in the first overlay packet, a source Internet Protocol (IP) address of the first service request packet is an IP address of a client, and a destination IP address is a public IP address associated with a virtual machine; the border gateway decapsulates the first overlay packet to obtain the first service request packet, encapsulates the first service request packet to generate a second overlay packet, and sends the second overlay packet to a virtual forwarding device, implementing transparent transmission of a source address of a service request packet.