Adversarial Variational Inference and Learning in Markov Random Fields.

摘要: Markov random fields (MRFs) find applications in a variety of machine learning areas, while the inference and such models are challenging general. In this paper, we propose Adversarial Variational Inference Learning (AVIL) algorithm to solve problems with minimal assumption about model structure an MRF. AVIL employs two variational distributions approximately infer latent variables estimate partition function, respectively. The distributions, which parameterized as neural networks, provide negative log likelihood On one hand, is intuitive form approximate contrastive free energy. other minimax optimization problem, solved by stochastic gradient descent alternating manner. We apply various undirected generative fully black-box manner obtain better results than existing competitors on several real datasets.