Artificial Intelligence for rent

© Fraunhofer IPA, Photo: Rainer Bez

© Fraunhofer-Gesellschaft

To be able to use artificial intelligence, a company does not necessarily need an expert. A Fraunhofer study shows how small and medium-sized enterprises can proceed instead.

Artificial Intelligence (AI) helps optimize production processes and save money. However, small and medium-sized enterprises often lack the expertise to utilize this future technology. They can collect the necessary data but struggle with its analysis. Here, large cloud providers help. They offer simple digital tools that process large datasets and deliver AI solutions. Experts refer to these as "Machine Learning-as-a-Service Platforms." This way, any company can get started with artificial intelligence without extensive experience and develop models that, for example, automatically detect defective workpieces.

Comparison of the most common use cases on four platforms

But which platform is suitable for which task? The Fraunhofer Institutes for Production Technology and Automation IPA and for Work Science and Organization IAO in Stuttgart compared the approaches of the four largest providers – AWS, Google, IBM, and Microsoft. They implemented solutions for four use cases that frequently occur in practice and cover four categories of data: tabular data, text, images, and time series:

- Customer churn: For hotels, it is advantageous to know early which guests are at risk of canceling. There may already be a hint in the tabular booking data. AI can detect it and develop an appropriate algorithm.
- Text categorization: Texts can be assigned to different categories, such as culture, sports, and politics. For example, a press agency can automatically maintain an archive.
- Image recognition: In production, image analysis plays an important role. Cameras can detect defects on the workpiece. AI helps automate this inspection. From a large number of annotated images with metadata, the AI learns to recognize errors.
- Tool wear: Replacing a milling cutter at the right time saves money. Acting too early wastes material, while acting too late risks long production downtime. AI learns to interpret the time series data of vibrations and power consumption to accurately assess the condition of the cutter.

In general, AI solutions follow the rule: the more data available and the better the quality of the data, the more reliable the resulting model. When comparing the platforms, the Fraunhofer scientists always chose the most accessible solution path. Often, it was only necessary to upload datasets and annotate them: in image processing, for example, by labeling each image as correct or defective. The platform then provided the desired model along with its prediction accuracy.

Results

The Fraunhofer study found that solutions from all providers demonstrate strengths and do not require deep expertise. Of course, there are some differences. Some platforms are more intuitive to use than others. Also, some AI models only run on the provider’s cloud, while others can be exported and installed on company-owned servers.

The study "Cloud-based AI Platforms – Opportunities and Limits of Services for Machine Learning as a Service" presents which platform is recommended for which use case. It is available for download at the following link: https://www.ki-fortschrittszentrum.de/de/studien/cloudbasierte-ki-plattformen.html