Biomedical Engineering Reference
In-Depth Information
not essential. It does not, however, describe any models of adaptation, learning,
evolution, etc. Many of those topics are covered in Flake's topic, which however
is written at a much lower level of mathematical sophistication.
On foundational issues about complexity, the best available surveys
(10,195) both neglect the more biological aspects of the area, and assume ad-
vanced knowledge of statistical mechanics on the part of their readers.
9.2. Data Mining and Statistical Learning
There are now two excellent introductions to statistical learning and data
mining: (223) and (31). The former is more interested in computational issues
and the initial treatment of data; the latter gives more emphasis to pure statistical
aspects. Both are recommended unreservedly. Baldi and Brunak (95) introduce
machine learning via its applications to bioinformatics, and this may be espe-
cially suitable for readers of the present volume.
For readers seriously interested in understanding the theoretical basis of
machine learning, (224) is a good starting point. The work of Vapnik
(22,225,226) is fundamental; the presentation in (22) is enlivened by many
strong and idiosyncratic opinions, pungently expressed. (40) describes the very
useful class of models called "support vector machines," as well as giving an
extremely clear exposition of key aspects of statistical learning theory. Those
interested in going further will find that most of the relevant literature is still in
the form of journals—
Machine Learning
,
Journal of Machine Learning Re-
search
(free online at www.jmlr.org),
Neural Computation
—and especially an-
nual conference proceedings—Computational Learning Theory (COLT), Inter-
national Conference on Machine Learning (ICML), Uncertainty in Artificial
Intelligence (UAI), Knowledge Discovery in Databases (KDD), Neural Informa-
tion Processing Systems (NIPS), and the regional versions of them (EuroCOLT,
Pacific KDD, etc.).
Much of what has been said about model selection could equally well have
been said about what engineers call
system identification
, and in fact
is
said in
good modern treatments of that area, of which (227) may be particularly rec-
ommended.
In many respects, data mining is an extension of exploratory data analysis;
the classic work by Tukey (228) is still worth reading. No discussion of drawing
inferences from data would be complete without mentioning the beautiful topics
by Tufte (229-231).
9.3. Time Series
Perhaps the best all-around references for the nonlinear dynamics approach
are (60) and (232). The former, in particular, succeeds in integrating standard
principles of statistical inference into the nonlinear dynamics method. (73),
while less advanced than those two topics, is a model of clarity, and contains an
