Maximization of ESI. Jaynes principle in testing significant inputs of linear model
A linear model \(z_i=\sum_{1\leq\lambda\leq t} b_\lambda x_i(\lambda)+e_i\) is considered, where \(z_i\), \(i=1,\dots,N\), are observed data, \(x_i(\lambda)\in\{-1,1\}\) are jointly i.i.d. random variables and \(e_i\) are i.i.d. random errors. The unknown coefficients \(b_\lambda\) are equal to 0 unless \(\lambda\in A\), where \(A\) is some set with cardinality \(| A| =s<<t\). The problem is to estimate the set \(A\). The authors propose two algorithms based on the empirical Shannon information. The first is asymptotically efficient but requires \(t^s\log t\) cycles of operations. The second one requires \(t\log t\) operations and is the best in some class of such tests. Both tests have mean error probability exponentially small in \(N\) as \(t\to\infty\).
- Optimum robust testing in linear models
- Maximal invariant likelihood based testing of semi-linear models
- Significance testing in non-sparse high-dimensional linear models
- Optimal testing for equicorrelated linear regression models
- A Bayesian decision theoretic interpretation of testing linear hypotheses
- A clarifying note on the effects of linear approximation on hypothesis testing
- Testing adequacy of linear random models
This page was built for publication: Maximization of ESI. Jaynes principle in testing significant inputs of linear model
Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q4231295)
