Towards Concept Based Software Engineering for Intelligent Agents
Meyer Ole, Volker Gruhn · 2019
The development of AI and machine learning applications at an industry mature level while maintaining quality and productivity goals is one of today's major challenges. Research in the field of intelligent agents has achieved many successes in recent years, especially due to various reinforcement learning techniques, and promises a high benefit in times of automation and autonomous systems. Bringing them into production, however, requires optimization against many other criteria than just accuracy. This leads to the emerging field of machine teaching. We already know many of the objectives used there from software engineering research, which has led to many well-established principles in recent decades. One of them is the component-based development whose idea finds an interesting counterpart in hierarchical reinforcement learning. We show that both areas can benefit from each other and introduce our approach of Concept Based Software Engineering, which is focused on supporting productivity and quality goals during the development of such systems.