Study about the learning algorithm applicable to large-scale Markov Random Fields

点击翻译按钮获取中文摘要

Learning and inference algorithms of the high-dimensional Markov Random Fields were developed for image processing and speech processing. In these application fields, Markov property is formed locally in spatial or temporal aspects. Group-coordinate descent was proposed for the optimization to exploit this locality. Since it enables to optimize the optimization problem without losing the convergence rate regardless of the dimensionality, it is applied to the real world problems such as image super-resolution and X-ray computed tomography.