Biomedical Engineering Reference
In-Depth Information
therefore, the ability of holistic judgment. Clearly, this capability cannot be implemented
in terms of sequential algorithm, as Conrad repeatedly pointed out. However, there is rea-
son for optimism in the future. An effective approach is again imitating Nature at the prin-
ciple level rather than outright imitation.
Decision-making in digital computers depends on logical operations of numerous discrete
flip-flop circuits, while decision-making in the human brain depends on a holistic process of
evaluating the interactions of numerous synapses at the neuronal level and arbitration of
conflicting factors at the cognitive level. The idea led to the conception of artificial neural
network computing, implemented in the environment of digital computers, that is, it is a vir-
tual machine within a digital machine. Artificial neural network research is among the
fastest growing science and technological endeavors, since Hopfield [144] published his
seminal article in 1982. Equally explosive is the growth of the literature of artificial neural
network; it defies a serious review here (see a highly readable review by Churchland [145]).
It suffices to cite a single example of its application to sensor technology.
Among all five human (special) senses, primary detection in the visual sense is rela-
tively straightforward. The interpretation of color in humans is based on the relative
degrees of activation of red, blue, and green (cone-shaped) photoreceptors (essentially
Helmholz's trichromatic theory). The process is somewhat discrete and algorithmic. That
is not to say human's recognition of visual forms and motion is also straightforward. In
contrast, the chemical sense provided by the olfactory (smell) and the gustatory (taste) sys-
tems is somewhat difficult to articulate in words and to simulate with computer algorithm
because differentiation of different smells or tastes depends more heavily on holistic judg-
ment than color discrimination does. While it was feasible to construct different chemical
sensors with a wide variety of specificities, criteria of differentiating a vast number of
volatile organic chemicals cannot be readily specified with a set of explicit rules or sequen-
tial computer algorithm.
Olfaction provides two categories of physiological functions. One is general odor sens-
ing, whereas the other has important behavioral consequence through sensing of
pheromones [146]. It turned out that general odor sensing and pheromone sensing are seg-
regated in different neural pathways. Even in general odor sensing, chemical structures
offer few clues. A slight change in chemical structures sometimes leads to a dramatic shift
in odor. Even a change of concentration can turn a pleasant odor into an offensive one and
vice versa (smelling of a skunk's secretion is a common experience).
The advent of modern molecular biology allowed a detailed dissection of olfactory
receptor molecules. The olfactory receptor molecules are encoded with multiple genes
[147]. While these receptor molecules are discrete, the sheer large number of them trans-
forms discrete sensing, as in color discrimination, into a
virtual continuum
, thus making
holistic judgment (or analog pattern recognition) possible. Nature's approach parallels the
transformation of discrete flip-flop operations into complex synaptic interactions. Of
course, comprehension of this vast continuum of information must be implemented at the
architecture level, as advocated by Conrad [125]. (So do visual comprehension of forms,
motion, and other high-level cognition [148]). Sometime ago, Pearce [149,150] relied on an
artificial neural network program to accomplish this feat (Artificial Electronic Nose).
Essentially, information is stored, in a distributed sense, among numerous “synapses” of
the network, much like what transpires in human's nervous system. Given a sufficient
number of “synapses,” discrete properties of individual synapses also coalesce into a vir-
tual continuum, thus converting a basically digital process into an equivalent analog
process. Again, this is a testimonial of the power of analog pattern recognition.
Another exciting development in nontraditional computing was what is now known as
DNA computing. Adleman [151] used standard tools of molecular biology to solve a case
of the directed Hamiltonian path problem—the so-called “traveling salesman” problem.
