A minimax approach to errors-in-variables linear models (Q1631206)

The author considers a simple linear regression model with Gaussian noise in both variables (regressor and regressand). The goal is to estimate the two unknown regression parameters. The classical maximum likelihood estimates show an unstable behaviour for large noise levels, e.g. unbounded moments. To overcome this disadvantage the author introduces a minimax approach and finds estimates with much more better behaviour under large noise regime. To construct these estimates a fast algorithm is proposed which solves the underlying convex optimization problem.