SAP offers two different types of functions within the Leonardo Machine Learning Foundation.
The Machine Learning Predictive Services provide functions for which, as usual in machine learning projects, a data set as representative as possible must first be provided, with which the system trains a model, i.e. learns, in order to then be able to use the following functions, among others:
clustering service
Clustering is used to form groups of the same type. With SAP Clustering Service, you can, for example, quickly segment customers or automatically sort articles into the right merchandise categories.
Forecast Service
The Forecast Service forecasts future numerical values based on previously known data. However, this is not just a simple extrapolation. More unknown factors are identified in the data, with which, for example, sales planning can be carried out more robustly or workload planning more specifically than with purely statistical methods.
Outliner Service
With outliner analysis, you can find outliers in data that would sink in the crowd. This can be used, for example, to analyze the material master for possible maintenance errors or to detect unusual ordering behavior on the part of customers in order to prevent potential fraud.
Recommendation Service
If product A is purchased, product B is also purchased with an X-% probability. These findings can be generated quickly and easily with SAP's Recommendation Service and do not have to be manually adjusted constantly and laboriously due to the continuous learning of the model. These functions have been successfully used by the very large online retailers for several years and are now available to companies of any size in any scalability quickly and easily via SAP Services.
All of the above functions are based on the company's own historical data that is used to create a model. In contrast, the functions within SAP Machine Learning Functional Services use models that have already been prepared by SAP and that do not first have to be trained, but can be used immediately via an API call. The following functions, among others, can be used within a very short time.
Image Classification and Feature Extraction API
With the Image Classification API, images can be evaluated and classified quickly and easily using a pre-trained SAP algorithm. This service can be used, for example, to obtain labels that enrich product images with keywords. Images can also be analyzed for similarity to find duplicates.
Topic Detection API
The Topic Detection API extracts topics from written text that the text deals with. For example, documents can be analyzed to better structure internal knowledge bases or keywords can be added to documents to make it easier to search for them.
Optical Character Recognition (OCR) API
SAP's OCR API recognizes text in PDF or image files and extracts it. Optical text character recognition is already frequently used in highly specialized applications, but mostly as a proprietary solution. With this API you can, for example, transcribe scanned incoming documents such as invoices or general letters and then use the Topic Detection API to extract the associated topics in order to keyword the documents.
Product Text Classification API
With the Product Text Classification API, SAP provides a service that analyzes product texts and automatically classifies them into predefined groups. This function can be used, for example, to assign a large number of articles to product groups or other product categories fully automatically and based on their article texts.
With the Machine Learning Functional Services, SAP offers an easy introduction to the topic of machine learning, enabling you to implement your first quick wins as quickly as possible. The pre-trained models drastically shorten the time to benefit, but are limited to specific applications and do not always take individual needs into account.
If company- or industry-specific questions are to be answered using Machine Learning, Machine Learning Predictive Services are better suited, as they can be individually adapted to the individual application. However, the effective and efficient use of these services goes hand in hand with increased effort in terms of data preparation, model creation and regular further development of the machine learning model.