The Greedy Top-Down Method of Experiment Selection

When the Shannon entropy (section 3.2) is used as uncertainty function,

$$I(X;S) = I(S;X)\protect$$

$$= H(X)-H(X\vert S)\textrm{,}$$ (53)

and $H(X)$ is, therefore, sometimes used as an approximate guide to the worth of an experiment $X$, the approximation being equivalent to an assumption that theories with significant prior probabilities each predict one result for the experiment with near certainty, and that $H(X\vert S)$ is therefore small; MacKay [44], in the contexts of simple measurement problems, and of the construction of codes for data compression, has named this approximation the `greedy top-down method.'

The author's description [30] of the motivation, for polarized electron reflection experiments on $Mn/Co/Cu(001)$ multi-layers, in terms of the continuing uncertainty over the magnetic moments of manganese atoms in these structures, is an example of a natural-language version (section 5.3) of a greedy top-down experiment selection method.