Extra Seminar Computer Science - Dr. V. Degeler

Title: Reasoning in collaborative smart environments for autonomous decision making and human guidance

Abstract:
Pervasive context-aware systems use distributed ubiquitous sensors to understand the environment and adapt it according to occupants' goals and preferences. In this talk, I'll discuss several different aspects of smart environments and their interaction with people. In active smart environments, the autonomy and reasoning power should be counterbalanced by the ability of people to understand and control the system's actions. The behavior of the system can be represented as a dynamic constraint satisfaction problem (CSP). The dependency graph (DG) allows not only to reduce CSP search space for every consecutive sensor event but also to give information about the exact reasons for the system’s decisions. The additional challenge is the complexity of obtaining enough information about the environment to correctly and fully assess the existing situation. Context Consistency Diagrams (CCD) are used to effectively represent different interpretations of the current situation, given possibly erroneous sensor readings. Collaborative smart systems assume the possibility to involve people in the process of making decisions, asking them to provide missing information or to perform actions that the system cannot perform itself. Such systems operate well in restricted partially observable environments. Smart automated systems must also be able to detect anomalous and potentially dangerous activities and to find the root causes of manifested danger signals. We will discuss a domain-independent danger explanation system that correlates anomalous activities with manifested danger signals to minimize the mislabelling of legitimate new behavior as dangerous in order to avoid human operators' alert fatigue. Finally, we will look at the reasoning of smart systems in the transportation domain. Traditional decision-making factors, e.g. trip duration, cost, transfers, are inadequate in many cases, as people also take into account their comfort, perception of reliability or safety, the state of mind, and familiarity with potential options. We will discuss, on a use case of bunching, how incorporation of these factors into recommender systems can improve not only people's comfort but also operations of the public transport system.