Sap Se

Sap Ag

Siar Sarferaz

Volker Wiechers

Bernd Schmitt

Markus A Peter

Bernhard Fippel

Joerg Wiederspohn

A computer-implement method can trigger deployment of an intelligence scenario on a tenant computing system. The intelligence scenario includes one or more software artifacts for a machine learning model and predefined metadata of the machine learning model. The method can automatically evaluate application data of the tenant computing system. Responsive to finding that the application data satisfies one or more requirements specified by the metadata, the method can automatically train the machine learning model using the application data and evaluate performance of the machine learning model trained using the application data. Responsive to finding that the performance of the machine learning model meets one or more performance criteria specified by the metadata, the method can automatically activate the machine learning model on the tenant computing system.

Artificial intelligence support in ERP systems

Techniques and solutions are described for facilitating the use of machine learning techniques. In some cases, a system suitable for providing a machine learning analysis can be different from a remote computer system on which training data for a machine learning model is located. A machine learning task can be defined that includes an identifier for at least one data source on the remote computer system. Data for the at least one data source is received from the remote computer system. At least a portion of the data is processed using a machine learning algorithm to provide a trained model, which can be stored for later use. Data on the remote computing system can be unstructured or structured. Particularly in the case of structured data, a remote computer system can make updated data available to the machine learning task.

Facilitating machine learning using remote data

Reducing resource consumption of a database system and a machine learning (ML) system is described. Data is received from a ML application of a database system. The data includes a first inference call for a predicted response to the received data. The first inference call is a request to a ML model to generate one or more predictions for which a response is unknown. A ML model using the received data generates an output comprising the predicted response to the data. The output for future inference calls is cached in an inference cache so as to bypass the ML model. The generated output to the ML application is provided by the ML model. A second inference call is received which includes the data of the first inference call. The cached output is retrieved from the inference cache. The retrieving bypasses the ML model.

Machine learning performance and workload management

Systems, methods, and computer program products are described herein for managing a lifecycle of a machine learning (ML) application from a provider point of view. Within a data intelligence platform, a package having ML scenarios and a training pipeline is generated. The training pipeline includes training logic associated with a defined workflow for training the ML application. The data intelligence platform is synchronized with a first database via an application programming interface. The first database generates a transport request containing the package. The transport request facilitates publication of content from the ML application. The ML application is assembled from the transport request within a second database. ML content is displayed on a graphical user interface associated with the second database.

Machine learning lifecycle management

Techniques and solutions are described for facilitating the use of machine learning techniques. In some cases, filters can be defined for multiple segments of a training data set. Model segments corresponding to respective segments can be trained using an appropriate subset of the training data set. When a request for a machine learning result is made, filter criteria for the request can be determined and an appropriate model segment can be selected and used for processing the request. One or more hyperparameter values can be defined for a machine learning scenario. When a machine learning scenario is selected for execution, the one or more hyperparameter values for the machine learning scenario can be used to configure a machine learning algorithm used by the machine learning scenario.

Facilitating machine learning configuration

Systems, methods, and computer program products for managing a lifecycle of a machine learning (ML) application from a consumer point of view are described herein. Execution of an intelligent scenario for training of the ML application is initiated. An integrator component generates a training pipeline. The training pipeline includes training logic associated with a defined workflow for the training. An application having an input dataset trains the ML application using the training pipeline. The integrator component determines training metrics associated with the trained ML application. The training metrics are indicators of a level of accuracy of the trained ML application. A centralized component provides the training metrics for characterization of the trained model.

A method may include training, based on a first training dataset, a machine learning model. A degradation of the machine learning model may be detected based on one or more accuracy key performance indicators including a prediction power metric and a prediction confidence metric. The degradation of the machine learning model may also be detected based on a drift and skew in an input dataset and/or an output dataset of the machine learning model. Furthermore, the degradation of the machine learning model may be detected based on an explicit feedback and/or an implicit feedback on a performance of the machine learning model. In response to detecting the degradation of the machine learning model, the machine learning model may be retrained based on a second training dataset that includes at least one training sample not included in the first training dataset. Related systems and articles of manufacture are also provided.

Detection of machine learning model degradation

Systems and methods are provided for receiving a request for data associated with a particular functionality of an application, identifying a first attribute for which data is to be generated to fulfill the request, and determining that the first attribute corresponds to data to be generated by a first machine learning model. The systems and methods further providing for executing a view or procedure to generate data for input to the first machine learning model, inputting the generated data into the first machine learning model, and receiving output from the first machine learning model. The output is provided in response to the request for data associated with the particular functionality of the application.

Embedded machine learning

Techniques and solutions are described for restricting data that is provided to a machine learning application. Restrictions can be based on use status information, such as use status information associated with a retention manager and indicating whether data is blocked from use. Data identifiers used by a cloud-based system can be correlated with archiving objects of a local system so that the cloud-based system can receive use status information to avoid using blocked data. Restrictions can include restricting data based on whether a data subject has provided consent that allows the data to be used by the machine learning application. A data view can be defined that filters query results to those where consent exits. The data view can join, such as an inner join, a table providing consent information with a data having data subject data.

Automatic enforcement of data use policy for machine learning applications

A framework for improving user interfaces, and in particular for improving user interfaces for displaying and interacting with predictive analytics, is described herein. In one embodiment, a user interface template renders predictive models and enables visually interacting with data to discover hidden insights and relationships in the data. The user interface template determines, based on the metadata and data annotations, how to display the supplied data. By encapsulating complex code necessary to render predictive models and enable visually interacting with data, the amount of frontend code required to implement predictive analytic functionality is reduced, defect rates are reduced, while design consistency is improved.

Smart template for predictive analytics

A method and system for generating an application has been described. A request is received to generate the application. Based on the received request, a Data Definition Language (DDL) query view defined for the application is executed to obtain a query view and a data transfer service. Next data is retrieved from the database based on the query view and the data transfer service. An application page template including a user interface (UI) related elements of the application is then generated based on the received request. The application page template and the retrieved data is then bound to generate a plurality of application pages of the application.

Generating applications based on data definition language (DDL) query view and application page template

Various embodiments herein each include at least one of systems, methods, and software for in memory data warehouse planning and broadcasting. Some embodiments include an in memory database having a set of define database table views that provide a virtual data model upon which services execute for various purposes including planning, simulation, and broadcasting of generated reports and other document. These services are executed within a computing environment of the in memory database and can be configured and grouped into applications and processes. Such embodiments eliminate system performance bottlenecks and provide a platform upon which 'extreme' application performance can be obtained.

In-memory data warehouse planning and broadcasting

The present disclosure relates to systems and methods for generating a virtual data model of data from one or more source systems, where the exact number of each type of source system is not known. Templates for each source system type may be defined and then explicit views may be generated for each source system based on configuration data provided by a customer.

Template based generation of cross views

A method includes processing a transaction on an in memory database where data being processed has a validity time, updating a time dependent data view responsive to the transaction being processed to capture time validity information regarding the data, and storing the time validity information in a historization table to provide historical access to past time dependent data following expiration of the validity time.

Growing community of inventors

Heidelberg, Germany

Siar Sarferaz