Shlomo,
As always, it's nice to here from you, and thanks for breaking the ice.
On Fri, 12 Oct 2001, Shlomo Geva wrote:
> Many of the target audience have extensive experience and pretty good
> understanding of pattern classification in the context of learning from a
> training set. There are many solutions (e.g ID3, C5, Neural Nets, Support
> Vector Machines, and many other.)
> ................................................................
We disagree on what the problem is all about, not what the solutions are.
> I have two questions, the answers to which may motivate people to study
> Evolving Transformation Systems (ETS):
> * Are ETS applicable to such problems?
Please keep in mind that any problem is formulated in the language of a
particular formalism, and the "real" problems can be formulated either
more or less adequately depending on the chosen formalism.
> * Is there published material that contains experimental results that
> show how ETS is applied and how well it performs on benchmark data sets.
I would prefer to answer "No", and one of the main motivations for this
discussion is to to move in this direction. However, having said this, its
only fair to add that a Ph.D. thesis (that is in the submission process)
by our group member John Abela has began to address your question.
Shlomo, more importantly, since you belong to "many of the target audience
[who] have extensive experience and pretty good understanding of pattern
classification in the context of learning from a training set", I would
like to ask you a relatively simple question. As we have heard many times,
the mankind is at the threshold of a new scientific era, the era of
understanding the nature of "intelligent" processes in the Universe.
Moreover, it is widely believed in the cognitive science community that
the pattern recognition processes are some of the most important cognitive
processes. Do you believe that "ID3, C5, Neural Nets, Support Vector
Machines, and many other" have ushered in this new scientific era? Which
non-trivial features (either formal or informal) of the intelligent
processes have they revealed? If none, how is it possible, then, to claim
their relevance to these (pattern recognition or inductive learning)
processes?
Cheers,
Lev
