Fusion ML and symbolic AI

Our motivation

Industrially deployed ML methods, such as Artificial Neural Networks or Deep Learning, make decisions in production every day. They support plant managers, machine operators, maintenance personnel and many more. It is not always clear how a process arrives at the respective decision. But especially in use cases where processes are to be understood and improved by humans, such insights offer new interesting ways to think and improve processes. In order to extract these insights from AI models, an explanatory interface for machine learning (ML) methods is being developed in the KOSMOX project. 

The aim is to make the decisions of the ML methods plausible and to communicate the findings to a broad mass of process experts, if possible in their respective technical language. Counterfactual explanation methods offer a promising approach here. These can be used and merged with the respective in-house processes to increase efficiency.

Our idea

In KOSMOX, a practicable interface is being created which, on the one hand, includes methods for explaining decisions made by ML procedures. On the other hand, this interface is also to be linked to internal company structures, so that employees without knowledge of data science can also derive direct added value from the use of ML processes.

Your benefit

Plant operators gain a deep understanding of the processes. Hidden correlations can be detected and used to eliminate defects and anomalies. The optimization of the process, for example the cycle time or the quality, is also made possible in a comprehensible way.

Partner

The project is carried out with the following partners:

• Fraunhofer IOSB-INA
• Deutsches Forschungszentrum für Künstliche Intelligenz GmbH (DFKI)
• ContiTech AG
• Lenze SE
• Villeroy & Boch AG