Developing methods for incorporating knowledge into machine learning
Jan. 01, 2021 to Dec. 31, 2023
In the »KI-Wissen« project, 16 project partners are working on artificial intelligence (AI)-based processes for fully automated driving. Until now, the development of such AI processes has been purely data-based. In the project, knowledge-based methods are now to be integrated to increase the functional quality, data efficiency, and plausibility, and validation of AI-supported functions.
The scientists at the Data Analytics Center (DANA) of Fraunhofer FOKUS are therefore working on integrating normative knowledge into the development. Specifically, this involves automating road traffic regulations using semantic AI methods and hybrid, sub-symbolic, and neurosymbolic integration with machine learning models. This enables normative knowledge, such as road traffic regulations, to be understood and complied with by AI components in the autonomous vehicle. Without such compliance with regulatory requirements, the approval of autonomous vehicles in the future is inconceivable.
At the same time, individual traffic situations occasionally require a deliberate, controlled break of the rules. For example, an obstacle, such as a deer on the road, may necessitate crossing a solid center line in areas where overtaking is prohibited. In such situations, it is also necessary to bring legal rules into harmony with general knowledge and acquired, unwritten rules of conduct and to decide on them. To this end, the DANA team is expanding the field of neuro-symbolic AI: Symbolic AI models used for semantic knowledge representation will be integrated with deep learning approaches (sub-symbolic AI models) and machine learning techniques for the detection and extraction of a structured scene, event, and situation information from autonomous vehicle data streams. In this way, semantic knowledge representation standards are further developed in the project and combined with machine learning.
»KI Wissen« is a project of the KI family and was developed from the VDA flagship initiative autonomous and connected driving. The project is funded by the German Federal Ministry for Economic Affairs and Energy (BMWi) based on a resolution of the German Bundestag.