Adaptive Inference on Probabilistic Relational Models

Standard approaches for inference in probabilistic relational models include lifted variable elimination (LVE) for single queries. To efficiently handle multiple queries, the lifted junction tree algorithm (LJT) uses a first-order cluster representation of a model, employing LVE as a subroutine in its steps. Adaptive inference concerns efficient inference under changes in a model. If the model changes, LJT restarts, possibly unnecessarily dumping information. The purpose of this paper is twofold, (i) to adapt the cluster representation to incremental changes, and (ii) to transform LJT into an adaptive version, enabling LJT to preserve as much computations as possible. Adaptive LJT fast reaches the point of answering queries again after changes, which is especially important for time-critical applications or online query answering.