Edward Liem: Optimizing Feature Causality and Blame


Event Details


Analyzing configurable systems using formal techniques, such as feature causality [1], allows for further insights towards why certain configurations exhibit a particular effect. Typically, the valid configuration space is exponential with respect to the number of features, and thus optimizations on feature cause computation would be beneficial. In the previous talk, we have discussed some promising optimization approaches. This talk will serve as a progress report on the findings and problems we have encountered, and which approach we have settled with for now.
Furthermore, we discuss the notion of Responsibility and Blame for feature causes. Adapted from Chockler and Halpern [2], we place a value on features to measure the degree of importance the feature has on producing the effect. Due to the exponential nature of the number of configurations and feature causes, we propose ideas to optimize the computation of these importance values. This is very much a work in progress and we will briefly present the idea behind our approach.

[1] C. Dubslaff, K. Weis, C. Baier, et al., “Feature causality”, Journal of Systems and Software, vol. 209, p. 111 915, 2024
[2] Chockler, H., Halpern, J.Y., 2004. Responsibility and blame: A structural-model approach. Artificial Intelligence Res. 22, 93–115.