An \(L_ 1\) smoothing spline algorithm with cross validation (Q1315221)

An algorithm for computing the LAD smoothing splines in the spaces \(W_ M(D)\), with \([0,1]^ n \subseteq D\), is presented. The data are assumed to be of the form \(y_ i = f(t_ i) + \varepsilon_ i\), \(1 \leq i \leq N\) with \(t_ 1,\dots,t_ N \in D\), the \(\varepsilon\)'s are errors with \(E(\varepsilon_ i)=0\), and \(f \in W_ M\). The LAD spline \(s_ \lambda (\lambda \geq 0)\) is defined as the solution of the minimization problem \[ \text{Min} \left \{(1/N) \sum^ N_{i=1} \bigl | y_ i-g(t_ i) \bigr | + \lambda J_ M (g):g \in W_ M \right\}, \] where \(J_ M(g)\) is a suitably chosen seminorm of \(g\). Also, the so-called LAD-CV smoothing spline algorithm is discussed. An application of the latter algorithm to a problem in environment data interpretation is included.