Biomedical Engineering Reference
In-Depth Information
from having an inadequate score function. This only emphasizes the
point that I made before, namely that it is much harder to develop a
good score function than it is to find out how to search for an
extremum of it. We will know that our score functions are good when
we find that re-optimization is no longer necessary.
Biology Buried in the Mathematics
The mathematics involved in optimization is relatively modest in a
technical sense but can, nevertheless, be quite complex and details
may be hidden from a user's eyes, for many reasons. My concerns
are:
Some algorithms, while wrapped in the guise of fulfilling a clini-
cally motivated prescription, are in fact mathematical formulae
with no biological basis or content
whatsoever.
The modest mathematical complexity of some models may obscure
both their physical and biological content or lack thereof.
Worst of all, the impenetrability of the algorithms may leave
some users no choice but to uncritically accept their conclusions.
When using automated optimization, it is the user's responsibility to
his or her patient to understand and be satisfied with what lies under
the covers, so to speak.
Caveat emptor
.
Figure 9.10. Hush. Optimization in progress.
