The automotive industry has started using Artificial Intelligence (AI) for autonomous vehicles, innovative user interfaces and predictive maintenance. AI can optimize huge numbers of parameters based on training data that would not be feasible without AI. Furthermore, there is and will in future be even more innovation in AI. For automotive SW development that means significant transformation effort to keep up. Key elements: Training: The performance of AI-based system depends heavily on training data. There are various sources of uncertainty that have a root cause in missing or unbalanced evidence of uncertain situations in the training data. But AI has advanced and there are solutions for this that must be adapted to automotive. For example, in natural language processing systems (e.g. ChatGPT), similar gaps in the training data are overcome by using task-agnostic unsupervised training on very huge data sets. Testing: With AI-based systems, there is a need to test with test data that is independent of the training data. Still the challenge of incomplete evidence due to missing data exists. AI Diligence: It is good practice to assess the development capability of automotive projects to ensure delivery in time, quality and budget. With AI, assessment models need to be adapted, focusing on the capability to apply new AI methods like avoiding over- and underfitting. In this talk we will explain key methods for AI training, testing and diligence and their practical application.
Session:
VEHICLE TECHNOLOGY II
|
| 12:30 - 13:00