Advances in Algorithms for Inference and Learning in Complex Probability Models

摘要: Computer vision is currently one of the most exciting areas artificial intelligence research, largely because it has recently become possible to record, store and process large amounts visual data. Impressive results have been obtained by applying discriminative techniques in an ad hoc fashion data, e.g., using support vector machines for detecting face patterns images. However, even more that researchers may be on verge introducing computer systems perform realistic scene analysis, decomposing a video into its constituent objects, lighting conditions, motion patterns, so on. In our view, two main challenges are finding efficient models physics scenes algorithms inference learning these models. this paper, we advocate use graph-based generative probability their associated analysis. We review exact various approximate, computationally techniques, including iterative conditional modes, expectation maximization algorithm, mean field method, variational structured Gibbs sampling, sum-product algorithm “loopy” belief propagation. describe how each technique can applied illustrative example multiple, occluding compare performances techniques.