Hidden Tree Markov Networks: Deep and Wide Learning for Structured Data

摘要

The paper introduces the Hidden Tree Markov Network (HTN), aneuro-probabilistic hybrid fusing the representation power of generative modelsfor trees with the incremental and discriminative learning capabilities ofneural networks. We put forward a modular architecture in which multiplegenerative models of limited complexity are trained to learn structural featuredetectors whose outputs are then combined and integrated by neural layers at alater stage. In this respect, the model is both deep, thanks to the unfoldingof the generative models on the input structures, as well as wide, given thepotentially large number of generative modules that can be trained in parallel.Experimental results show that the proposed approach can outperformstate-of-the-art syntactic kernels as well as generative kernels built on thesame probabilistic model as the HTN.