2017.31: Effective Condition Number Bounds for Convex Regularization
2017.31: Dennis Amexlunxen, Martin Lotz and Jake Walvin (2017) Effective Condition Number Bounds for Convex Regularization.
Full text available as:
|PDF - Requires a PDF viewer such as GSview, Xpdf or Adobe Acrobat Reader|
We derive bounds relating the statistical dimension of linear images of convex cones to Renegar's condition number. Using results from conic integral geometry, we show that the bounds can be made to depend only on a random projection to a lower dimensional space, and can still be effective if the linear maps are ill-conditioned. As an application, we get new bounds for the undersampling phase transition of composite convex regularizers. Key tools in the analysis are Slepian's inequality, interpreted as monotonicity property of moment functionals, and the kinematic formula from integral geometry. The main results are derived in the generalized setting of the biconic homogeneous feasibility problem.
|Item Type:||MIMS Preprint|
|Uncontrolled Keywords:||Convex regularization; compressed sensing; random matrix theory; Gaussian width; condition numbers|
|Subjects:||MSC 2000 > 60 Probability theory and stochastic processes|
MSC 2000 > 65 Numerical analysis
MSC 2000 > 90 Operations research, mathematical programming
|Deposited By:||Dr. Martin Lotz|
|Deposited On:||11 September 2017|