Dear
All,
It is
fascinating how sometimes seemingly totally different aspects of life like
engineering, science, human psychology and social behaviour come together as if
all have the same underlying cause and structure. Take for example human
behaviour, and global optimisation approaches which we use to solve many
problems of image processing and computer vision.
In global approaches in order to interpret
the data we usually have to minimise a cost function with two terms: the
faithfulness to the data term and the model term. I like to call the latter the
"prejudice about the world" term. The parameter that balances the two
is of such crucial importance, that it effectively dictates the solution. Make
that parameter heavy on the side of the model and you can produce any solution
you want. Make the parameter heavy on the side of the data and you had it!
For
example, have you ever wandered why people quarrel about things that to third
parties seem blatantly clear cut? Have you ever come across people who insist
on their views even when hard facts are presented to them about the opposite? I
find it incredible that two people may read the same facts but come to exactly
the opposite conclusions! This is because they use a model-based interpretation
of the data. They have to, otherwise the world does not make sense, just like
our data with no model are not interpretable. However, in model-based
interpretation, one has to remember that the question asked is: "Do the
data have a violent objection in being fitted by this model (my pet
model)?" Most of the data can be fitted by many models, particularly if we
bend them a little. Often the data obligingly answer: "No, I do not mind
if you try to fit me with your pet model!" The trouble starts when in the
excitement of the success and the narcissism of the
"pet-model-fits-wel1" mood, the exact nature of the question is
f`orgotten: The pet-model becomes religion, and takes the dimensions of THE
model that fitted the facts! Then somebody else comes along with his own prejudices
and loaded predispositions expressed by his own pet-model. He sees the same
facts, the same data and asks them: "Do you mind if' I fit you with my own
pet model my sweetie?" The data are obliging again: "Not at all! Suit
yourself!" Put these two ecstatic by their success and foresight people in
the same room, and you have a war! Am I exaggerating?
Maria
Petrou
PS: Beware of my own prejudicial model of the
world!