In the development of Machine Learning-based applications, tools play a prevalent role. This extends to the use of Machine Learning (ML) in flight-critical applications. ML-based tools, such as generative Artificial Intelligence (AI) tools, and traditional tools used for ML applications, like simulation generation tools, can accomplish various processes and objectives in the ML-based item lifecycle. These include creating high-level requirements (HLRs) and generating synthetic datasets used for model training, validation, and testing. The integration of tools for such applications raises crucial questions about tool qualification. Tool qualification criteria are traditionally contingent on the potential for a tool to introduce errors into its output that may impact the final deployed item. The examination of tool qualification further involves distinguishing between scenarios where the tool may overlook errors during testing. Higher tool qualification rigor is deemed necessary, particularly for the development of items with a higher assurance level. In the ML-based item lifecycle, evaluating tool qualification is vital, especially for the ML development lifecycle, because qualification increases the assurance of the tool and confidence in its output. This paper explores the applicability of tool qualification across the ML-based lifecycle and then examines two specific examples: the utilization of generative AI tools and synthetic data generation tools for ML training, validation, and testing.
Considerations for Tool Qualification in Flight-Critical Applications Using Machine Learning
2024-09-29
1512294 byte
Conference paper
Electronic Resource
English
Software tool qualification considerations
TIBKAT | 2011
|Hydraulic Fluid Qualification: An Overview of Critical Considerations
SAE Technical Papers | 2000
|Hydraulic fluid qualification: an overview of critical considerations
Automotive engineering | 2000
|