Discriminative Training of SumProduct Networks by Extended BaumWelch
[edit]
Proceedings of the Ninth International Conference on Probabilistic Graphical Models, PMLR 72:356367, 2018.
Abstract
We present a discriminative learning algorithm for SumProduct Networks (SPNs) \citep{poon2011sum} based on the Extended BaumWelch (EBW) algorithm \citep{baum1970maximization}. We formulate the conditional data likelihood in the SPN framework as a rational function, and we use EBW to monotonically maximize it. We derive the algorithm for SPNs with both discrete and continuous variables. The experiments show that this algorithm performs better than both generative ExpectationMaximization, and discriminative gradient descent on a wide variety of applications. We also demonstrate the robustness of the algorithm in the case of missing features by comparing its performance to Support Vector Machines and Neural Networks.
Related Material


