2018-11-02: Prof. Rüdiger Ehlers: Engineering Self-Monitoring and Self-Adaptive Systems
30.11.2018, 13:30 – 14:30 Uhr, Mühlenpfordtstraße 23, Raum 812 (Architekturtower)
The growing interest in self-adaptive systems raises the question of how their correct behavior can be assured. Such systems frequently employ machine learning
and other concepts from the area of artificial intelligence, which makes their reaction to the possible environment stimuli hard to predict. As a consequence,
giving correctness guarantees for safety-critical self-adaptive systems is difficult, which advocates for a rigorous approach to engineering them. This talk will provide an overview of two of my recent results on ensuring the correct behavior of systems that adapt to their environment. In the first part of the talk, I will show how to enforce that a safety-critical
and possibly self-adaptive system satisfies its specification without changing its implementation. The starting point is a new approach to monitoring that detects violations earlier than traditional trace-based monitoring and as early as theoretically possible. The new approach gives rise to the possibility to correct
the executed actions, which is particularly helpful for system components that employ machine learning. The second part of the talk concerns the question of how we can define reasonable behavior for deteriorating systems that continuously interact with their environment. A new control policy optimality notion gives rise to a method forcomputing risk-averse control policies, i.e., those that delay the failure of the system as long as possible. The resulting policies are suitable for fail-safe