Framework for Quality Assurance and Quality Control in DL Systems
Domingos F. Oliveira, Miguel A. Brito · 2023
The use of deep learning systems represents a significant challenge for the industry, as the quality assurance and control of these systems still need to be resolved, for the simple fact that these systems suffer from defects and vulnerabilities, which can lead to severe tragedies, mainly when applied to safety-critical applications in the real world, as they present a programming paradigm that differs in the representation of decision logic, as well as their development practices are unclear and centre on the data-driven paradigm, so how to ensure and control the quality of this system is unclear. This article discusses the development of a framework that will help specialists implement quality assurance and quality control in developing these systems. The framework combines two widely used tools, namely PDCA, applied to the CRISP-DM set of activities to enable quality assurance and control in developing DL systems, specifically in DS. In order to be implemented in the industry, it needs to be evaluated and validated. To this end, case studies and expert evaluation will be used to assess and validate the framework, as this stage will be addressed in future work.