CS: A Novel Algorithm for Compressive Sensing: Iteratively Reweighed Operator Algorithm, Machine Learning Summer School

Here is a new Iteratively Reweighted algorithm:

A Novel Algorithm for Compressive Sensing: Iteratively Reweighed Operator Algorithm (IROA) by Lianlin Li, Fang Li. The abstract reads:

Compressive sensing claims that the sparse signals can be reconstructed exactly from many fewer measurements than traditionally believed necessary. One of issues ensuring the successful compressive sensing is to deal with the sparsity-constraint optimization. Up to now, many excellent theories, algorithms and software have been developed, for example, the so-called greedy algorithm ant its variants, the sparse Bayesian algorithm, the convex optimization methods, and so on. The formulations for them consist of two terms, in which one is and the other is (, mostly, p=1 is adopted due to good characteristic of the convex function) (NOTE: without the loss of generality, itself is assumed to be sparse). It is noted that all of them specify the sparsity constraint by the second term. Different from them, the developed formulation in this paper consists of two terms where one is with () and the other is . For each iteration the measurement matrix (linear operator) is reweighed by determined by which is obtained in the previous iteration, so the proposed method is called the iteratively reweighed operator algorithm (IROA). Moreover, in order to save the computation time, another reweighed operation has been carried out; in particular, the columns of corresponding to small have been excluded out. Theoretical analysis and numerical simulations have shown that the proposed method overcomes the published algorithms.

I just saw the following: Machine Learning Summer School/Workshop on Theory and Practice of Computational Learning 2009 at University of Chicago include many topics that covers Compressive Sensing and related subjects. The Web page: http://www.cse.ohio-state.edu/mlss09/

2009 Summer School on Theory and Practice of Computational Learning and the associated workshop will be held at the University of Chicago from June 1 to June 11, 2009.
The summer school will cover abroad range of topics at a level appropriate for graduate students and researchers from other fields. The topics will include:

• Foundations of Statistical Learning
• Kernel Methods and Support Vector Machines
• Semi-supervised and Active Learning
• Boosting and Ensemble methods
• Compressed Sensing and Sparse representations
• Manifold Methods and Geometry of Point Clouds
• Graphical Models
• Machine Learning in Computer Vision, Speech, Text and Natural Language
• Processing
• Learning in Neuroscience and Human Computer Interaction

Workshop talks will be centered around similar subjects but at more advanced levels, reflecting the current research interests of the speakers. We plan to have approximately four hours of tutorial talks in the morning followed by four hours of workshops talks in the afternoon. This summer school/workshop structure will allow the participants to be introduced to the subjects of current interest and will also provide them with a unique educational opportunity to both learn the introductory material as well as to have access to a broad spectrum of research by some of the leading experts in the field.