Presented by: Mingyuan Zhou Duke University, ECE Feb 22, 2013 Large Scale Variational Bayesian Inference for Structured Scale Mixture Models Young Jun Ko and Matthias Seeger ICML 2012 Presented by: Mingyuan Zhou Duke University, ECE Feb 22, 2013
Introduction Natural image statistics exhibit hierarchical dependencies across multiple scales. Non-factorial latent tree models. A large scale approximate Bayesian inference algorithm for linear models with non-factorial (latent tree-structured) scale mixture priors. Experimental results on a range of denoising and inpainting problems demonstrate substantially improved performance compared to MAP estimation or to inference with factorial priors.
Structured Image Model Impose sparsity
Structured Image Model Non-factorial scale mixture model: Output:
Example
Large Scale Variational Inference Due to strong dependencies between components of u and s, factorial assumption might be restrictive. Iterative decoupling Decouple Decouple mean and covariance components of Prior:
Large Scale Variational Inference VB
Large Scale Variational Inference
Image denoising
Image inpainting
Conclusions