An Empirical Study of w-Cutset Sampling for Bayesian Networks
This work addresses efficiency in probabilistic inference for Bayesian networks, but it is incremental as it empirically studies an existing algorithm.
The paper investigates the time-space trade-off in w-cutset sampling for Bayesian networks, showing that restricting the cutset size to bound induced-width by w leads to fewer samples needed for convergence but increases per-sample inference time, with experiments demonstrating substantial benefits from optimal balance.
The paper studies empirically the time-space trade-off between sampling and inference in a sl cutset sampling algorithm. The algorithm samples over a subset of nodes in a Bayesian network and applies exact inference over the rest. Consequently, while the size of the sampling space decreases, requiring less samples for convergence, the time for generating each single sample increases. The w-cutset sampling selects a sampling set such that the induced-width of the network when the sampling set is observed is bounded by w, thus requiring inference whose complexity is exponential in w. In this paper, we investigate performance of w-cutset sampling over a range of w values and measure the accuracy of w-cutset sampling as a function of w. Our experiments demonstrate that the cutset sampling idea is quite powerful showing that an optimal balance between inference and sampling benefits substantially from restricting the cutset size, even at the cost of more complex inference.