Accountability

Speaker: Alexander Pretschner

Abstract

Accountability is the property of a system to help answer questions regarding why specific events have happened. With accountability infrastructures in place, identified reasons can be used to improve systems and to assign blame. Accountability rests on two pillars, monitoring and causality analyses. In this series of lectures, we focus on causality analyses and respective underlying models and show their wide applicability in computer science. We consider causality analyses including spectrum-based fault localization, Granger causality analysis, model-based diagnosis, and focus on SAT-based and ILP-based approaches to counterfactual reasoning on the grounds of Halpern and Pearl’s notion of actual causality inference. We also discuss the provenance of causal models as fault trees, attack trees, and explicit acyclic equations.