Fast And Accurate Learning of Probabilistic Circuits by Random Projections

Renato Geh, Denis Mauá

Keywords: Probabilistic circuits, structure learning

TL;DR: We show how to learn probabilistic circuits by random projections similar to decision trees.

Abstract: We revisit random projection trees in the context of probabilistic circuits. We show how a recursive partitioning scheme for inducing oblique kd-trees from data using random projections can be adapted to produce reasonably accurate probabilistic circuits in a fraction of the time used by typical structure learning algorithms.