Re: [Fis] The timings of meaning

R. Ulanowicz wrote:
> For the past 30 or so years I have been engaged in the application of
> IT to ecosystems flow networks. From the very outset I was struck by
> the want of any necessity to invoke the communications theory format
> of sender-
> channel- receiver. It appears that at least the Shannon formulation is
> applicable in situations so removed from the cognitive scene that some
> more general perspective on information must be possible -- much like
> Popper generalizes physical forces as "propensities".

Hello -

I'm a neophyte in this group, working on an approach to data analysis
inspired by the concepts from information theory, solving the kind of
problems where statistics and data mining are applied. Before that, I was
working in data compression (itself a branch of information theory),
participating in the development of the PNG and JPEG-LS imaging compression
standards, now both recognized by ISO.

Shannon's formulation can be seen as an application of a particular type of
analysis to the problem of communication. So I agree with Robert that unless
there is a good reason, one should not seek receivers, channels and senders.
Instead, the crucial components for analysis are attributes and entities.
Attributes are timeless aspects of reality that are always measured in the
same way, they are the concepts, the variables. Entities are temporally
and/or spatially delimited aspects of reality, restricted to a particular
zone or period. When we cross-link attributes and entities, we obtain
instances. Assume a medical example, where entities are patients and
attributes are various types of medical observations:

instance A.N.: temperature=40, tongue=yellow, age=old, eyes=white,
disease=fatal
instance M.O.: temperature=37, tongue=green, age=old, eyes=pink,
disease=none
instance B.D.: temperature=38, tongue=red, age=young, eyes=white,
disease=fatal
instance J.M.: temperature=37, tongue=red, age=medium, eyes=white,
disease=fatal

The set of these instances comprise the data. This is what one needs to
apply information theory. A sender S and a receiver R are both attributes
(assumed to be timeless), and the bits they exchange are entities (assumed
to be localized in time). Channel may also be an attribute, but more often
it is merely the interaction between the sender and the receiver,
mathematically described as a joint PDF P(S,R). However, more generally, the
object of study are the probability density functions on the attributes
while considering individual instances as samples. The model in the above
example would be P(temperature,tongue,age,eyes,disease), or if we
specifically wanted to predict the disease,
P(disease|temperature,tongue,age,eyes). The PDF is the matematical
formalization of the concept of probability, and propensity is a certain
view of probability.

We can approach this problem from two directions, the confirmational and the
exploratory. In the confirmational approach, we postulate these functions
and verify how well they fit the data. In the empirical approach, we try to
infer them from the data.

My own work is about identifying the structure of mutual and generalized
mutual information. The ordinary 2-way mutual information
I(temperature;tongue) is the difference between the holistic,
dependence-assuming P(temperature,tongue), and the reductionistic
independence-assuming P(temperature)P(tongue). The higher the mutual
information, the higher the possibility that there is something connecting
temperature and tongue color. It can be many things, it can be chance, an
embedded constraint, a boundary condition, a direct interaction, a material
cause, a formal cause, a final cause, an efficient cause. Due to Simpson's
paradox, we can never be sure.

BTW, L. Orloci is also working on applications of information theory to
ecology http://mywebpage.netscape.com/lorloci/koa

===
mag. Aleks Jakulin
http://ai.fri.uni-lj.si/aleks/
Artificial Intelligence Laboratory,
Faculty of Computer and Information Science, University of Ljubljana.

_______________________________________________
fis mailing list
fis@listas.unizar.es
http://webmail.unizar.es/mailman/listinfo/fis
Received on Wed Mar 31 09:30:15 2004