Abstract: According to implementations of the subject matter described herein, there is provided a solution for predicting the resource usage of the deep learning model. In this solution, information about a deep learning model is obtained, the information comprising first information for describing the deep learning model and second information about an operating environment of a job associated with the deep learning model. The static resource usage of the job is determined based on the first information and a strategy of the job during runtime in the operating environment is determined. Afterwards, resource usage of the job during runtime in the operating environment is predicted based on the strategy and the static resource usage. With this solution, the usage of various resources of the deep learning model, such as computation power consumption, memory consumption, execution time, and the like, under a specific runtime strategy can be accurately predicted.

Abstract: Implementations of the present disclosure relate to visual programming for deep learning. A computer-implemented method comprises presenting a visual representation of an artificial neural network, the visual representation comprising graphical elements representing layers of the artificial neural network; in response to receiving a drag-and-drop operation on the graphical elements, modifying an intermediate representation of the artificial neural network, wherein the intermediate representation is independent of a deep learning framework and the drag-and-drop operation is configured to modify connections between the graphical elements; and modifying, based on the intermediate representation of the artificial neural network, code of the artificial neural network for a target deep learning framework.

Abstract: Data-parallel computation programs may be improved by, for example, determining the functional properties user defined functions (UDFs), eliminating unnecessary data-shuffling stages, and/or changing data-partition properties to cause desired data properties to appear after one or more user defined functions are applied.

Abstract: Data-parallel computation programs may be improved by, for example, determining the functional properties user defined functions (UDFs), eliminating unnecessary data-shuffling stages, and/or changing data-partition properties to cause desired data properties to appear after one or more user defined functions are applied.

Abstract: Program partitioning of an application can include creating execution flow graphs and static flow graphs of targeted functions or operations of the application. Based on the execution flow graphs or static flow graphs, replay interfaces are created. The replay interfaces provide data flows that are usable in re-execution of the application during program development.

Abstract: Analyzing and detecting soft hang program errors may lead to suggestions for either curing the programming errors at runtime or refactoring the source code. For instance, responsive function invocation patterns and blocking function invocation patterns may be used to detect soft hang program errors in a source code file. Deductive database rules may be compiled from the responsive and blocking function invocation patterns to find matching function invocations in a call graph.

Abstract: Described techniques increase runtime performance of software running in user space by analyzing and detecting soft hang program errors and giving suggestions for cures. This disclosure pertains to techniques for the analysis, detection, and cure of soft hang program errors.