Greedy Part-Wise Learning of Sum-Product Networks

Abstract

Sum-product networks allow to model complex variable interactions while still granting efficient inference. However, most learning algorithms proposed so far are explicitly or implicitly restricted to the image domain, either by assuming variable neighborhood or by assuming that dependent variables are related by their magnitudes over the training set. In this paper, we introduce a novel algorithm, learning the structure and parameters of sum-product networks in a greedy bottom-up manner. Our algorithm iteratively merges probabilistic models of small variable scope to larger and more complex models. These merges are guided by statistical dependence test, and parameters are learned using a maximum mutual information principle. In experiments our method competes well with the existing learning algorithms for sum-product networks on the task of reconstructing covered image regions, and outperforms these when neither neighborhood nor correlations by magnitude can be assumed.