Abstract: This application discloses a model training method and apparatus based on a hybrid parallelism manner. In this method, a neural network model is divided into a plurality of pipeline stages, and each pipeline stage includes a plurality of sub-stages of the neural network model. Computing nodes corresponding to the plurality of pipeline stages are invoked in a hybrid parallelism manner according to a sequence of sub-stages in the neural network model. When iterative training is performed on a network layer in a corresponding pipeline stage, because sub-stages at same locations in adjacent pipeline stages are consecutive in the neural network model, the computing node does not need to wait for completion of forward propagation of a previous pipeline stage, and can perform forward propagation on the corresponding pipeline stage only after forward propagation of the 1st sub-stage in the previous pipeline stage is completed.

Abstract: Embodiments of this application provide application processing methods and apparatuses. One method includes: receiving a startup request from a first client device, where the startup request is used to start an application. Loading, in a high-level language execution environment of an enclave, a manifest file of the application and a dependency relationship between an enclave entrypoint function of the application and a static dependency class, and loading a static dependency class of the enclave entrypoint function of the application based on the manifest file and the dependency relationship between the enclave entrypoint function and the static dependency class.

Abstract: Embodiments of this application provide application processing methods and apparatuses. One method includes: receiving a startup request from a first client device, where the startup request is used to start an application. Loading, in a high-level language execution environment of an enclave, a manifest file of the application and a dependency relationship between an enclave entrypoint function of the application and a static dependency class, and loading a static dependency class of the enclave entrypoint function of the application based on the manifest file and the dependency relationship between the enclave entrypoint function and the static dependency class.

Abstract: Disclosed are a data processing method and apparatus, a computing device, and a storage medium, which belongs to the field of big data technologies. The method includes: determining, based on a query algorithm, a first query output (601) corresponding to an input data set; determining, based on a second query output of an unsampled data record in the input data set and the query algorithm, a perturbation of each of a target quantity of sampled data records in the input data set to the first query output, to obtain sensitivity (602) corresponding to the input data set, where the unsampled data record and the target quantity of data records constitute the input data set; and adding noise to the first query output based on the sensitivity and outputting a noised first query output (603).

Abstract: Methods, systems, and media for providing determinism in multithreaded programs are provided. In some embodiments, methods for providing determinism in multithreaded programs are provided, the methods comprising: instrumenting a program with: a proxy that: receives inputs to the program; and determines whether a schedule corresponding to a set of input constraints satisfied by the inputs has been memoized; and a replayer that: if it is determined that a schedule corresponding to a set of input constraints satisfied by the inputs has been memoized, processes the inputs by the program according to the schedule.

Abstract: Methods, systems, and media for protecting applications from races are provided. In some embodiments, methods for protecting applications from races are provided, the methods comprising: adding to at least one cycle of the application an update check to determine when an update to the application is to be made; adding an update engine to the application, wherein the update engine is configured to: receive an update plan that is based on an execution filter that specifies how operations of the application are to be synchronized; and cause synchronization operations to be added to the application based on the update plan that prevent the race from occurring.

Abstract: Methods, systems, and media for protecting applications from races are provided. In some embodiments, methods for protecting applications from races are provided, the methods comprising: adding to at least one cycle of the application an update check to determine when an update to the application is to be made; adding an update engine to the application, wherein the update engine is configured to: receive an update plan that is based on an execution filter that specifies how operations of the application are to be synchronized; and cause synchronization operations to be added to the application based on the update plan that prevent the race from occurring.

Abstract: Methods, systems, and media for providing determinism in multithreaded programs are provided. In some embodiments, methods for providing determinism in multithreaded programs are provided, the methods comprising: instrumenting a program with: a proxy that: receives inputs to the program; and determines whether a schedule corresponding to a set of input constraints satisfied by the inputs has been memoized; and a replayer that: if it is determined that a schedule corresponding to a set of input constraints satisfied by the inputs has been memoized, processes the inputs by the program according to the schedule.

Abstract: Methods, systems, and media for protecting applications from races are provided. In some embodiments, methods for protecting applications from races are provided, the methods comprising: adding to at least one cycle of the application an update check to determine when an update to the application is to be made; adding an update engine to the application, wherein the update engine is configured to: receive an update plan that is based on an execution filter that specifies how operations of the application are to be synchronized; and cause synchronization operations to be added to the application based on the update plan that prevent the race from occurring.