A Geometric View on Constrained M-Estimators
arXiv:1506.08163
Abstract
We study the estimation error of constrained M-estimators, and derive explicit upper bounds on the expected estimation error determined by the Gaussian width of the constraint set. Both of the cases where the true parameter is on the boundary of the constraint set (matched constraint), and where the true parameter is strictly in the constraint set (mismatched constraint) are considered. For both cases, we derive novel universal estimation error bounds for regression in a generalized linear model with the canonical link function. Our error bound for the mismatched constraint case is minimax optimal in terms of its dependence on the sample size, for Gaussian linear regression by the Lasso.
References in corpus (12)
- Simultaneous analysis of Lasso and Dantzig selector
- Quantum state tomography via compressed sensing
- Quantum Tomography via Compressed Sensing: Error Bounds, Sample Complexity, and Efficient Estimators
- A Unified Framework for High-Dimensional Analysis of M-Estimators with Decomposable Regularizers
- Computational and Statistical Tradeoffs via Convex Relaxation
- Compressed sensing performance bounds under Poisson noise
- A new perspective on least squares under convex constraint
- High-dimensional estimation with geometric constraints
- Exponential Family Matrix Completion under Structural Constraints
- The generalized Lasso with non-linear observations
- Estimation in high dimensions: a geometric perspective
- Guarantees of Total Variation Minimization for Signal Recovery