Statistical Runtime Monitoring of Adaptive Systems

Speaker: Konstantin Kueffner, University of Oxford
Title: Statistical Runtime Monitoring of Adaptive Systems
Abstract: The autonomous agents that increasingly permeate modern societies give rise to adaptive systems: their actions alter their environments, and they adapt to the resulting changes, creating continuously evolving systems whose failures can harm people and critical infrastructure. Formal verification aims to prevent such harm by certifying, traditionally before deployment, that a model of the system satisfies a behavioural specification. The resulting guarantees therefore depend on the alignment between the model and the system. For adaptive systems, however, this correspondence may deteriorate as the system evolves, eroding the established guarantees. My work aims to detect this erosion with statistical runtime monitoring. In this talk, I will use examples from monitoring algorithmic fairness and supermartingale certificates to show how observations of a running system can reveal when the guarantees no longer hold, by checking either the property itself or the assumptions underlying its proof.
Join: https://teams.microsoft.com/meet/3784779010650946?p=mjVR4sSoILNMhQgLAD
We ask that you please familiarise yourself with the Hub’s code of conduct prior to attendance. Please note, this talk will be recorded.
