The exponential growth in data, higher-resolution imaging, and expanding wireless communication highlight the importance of sparse models for efficient data representation and recovery. This thesis presents a novel algorithm for fast sparsifying transforms applicable to arbitrary transformation matrices and adapts existing compressed sensing algorithms for sparse recovery from heavy-tailed measurements. The key innovation is the use of a median-of-means estimator, providing robustness against outliers and ensuring strong concentration results.
«
The exponential growth in data, higher-resolution imaging, and expanding wireless communication highlight the importance of sparse models for efficient data representation and recovery. This thesis presents a novel algorithm for fast sparsifying transforms applicable to arbitrary transformation matrices and adapts existing compressed sensing algorithms for sparse recovery from heavy-tailed measurements. The key innovation is the use of a median-of-means estimator, providing robustness against ou...
»