FYI, I recently received the advice below. Accordingly, I've changed
our description to "task force".

This got me thinking - "task force" sounds somehow more dynamic than
"working group". What do you all think about ramping up our
activities? A quick search on google shows up a GECCO workshop in 2002
and a special session at GECCO in 2003. The call for special sessions
for WCCI 2008 is still open. What about a special issue of TEC?

Any volunteers?

cheers, phi

***********

Dear ECTC WG chair

The ECTC members have voted to amend our charter so that ECTC would
have "Task Forces" rather than "Working Groups".

Here is the reasoning. All other CIS committees have task forces, and
their definition is similar to what we define as working groups.  This
has become a bit annoying when ECTC presents alongside other
committees, and the CIS President has asked me to raise this with the
committee. Two notes of clarification. First, WG and TF definitions
were discussed extensively by the committee before approving them at
the time, and I still believe they are the most useful to have.
Second, most useful or not, it would be rather hard - and also a waste
- to convince all other committees of our point of view.

I would now ask you to do the following please:

1. Use Task Force for your group and amend any literature entries
accordingly.

2. Provide me with a list of membership of your task force, including
affiliation. This is requested by the CIS President for their records.

Please reply within a week so I can feedback to Vin in good time.

Best wishes.

Ali

--
Ali M.S. Zalzala
PhD, CEng, SMIEEE, MIEE

CEO
Hikma Group Limited, P.O. Box 42423, Dubai, UAE
Tel: +971 (0)50 6530372. Fax: +971 (0)4 2246046. Email:
ad...@hikmagroup.com. Web: www.hikmagroup.com. Personal Web:
www.zalzala.info
--
This e-mail is from Hikma Group Limited. The information in this e-
mail and any files transmitted with it are confidential and may be
legally privileged. It is solely for the addressee intended by Hikma
Group Limited and access by any other person to this e-mail is not
authorized. If you are not the intended recipient, please notify to us
immediately then please delete this email and confirm to us. Any
unauthorized use or disclosure of this e-mail, or of the information
contained therein or any copying, distribution, dissemination of it is
prohibited, and may be illegal.

Our address is Hikma Group Limited, P.O. Box 42423, Dubai, UAE. Fax
no:
+971.4.2246046 .  Email: ad...@hikmagroup.com

Hikma Group Limited and its affiliates do not accept any liability for
any errors, omissions, corruption or viruses contained within this e-
mail or within any files transmitted with it, or which may arise as a
result of its transmission. Any views expressed by an individual
within this e-mail do not necessarily reflect the views of Hikma Group
Limited or its affiliates.

I like the idea of fostering community, and I do not oppose your
suggestions; however, I wonder what the role of this Google Group is,
can be, or should be.  Perhaps a broader scope-and-goal oriented
conversation would be useful.

One might argue that the kind of community you mention is arising
naturally.  There is certainly an active, though I believe quite
small, community of researchers studying coevolutionary computation
right now.  Moreover, I think that in recent years, there's not only
been increased activity, but an increase in relationships between
different groups studying coevolution (the Brandeis group and the GMU
group certainly have stronger links now than they did five or so years
ago, for example).

And the activity has shown at the conferences:  Not just the theory
workshop and the discussion forum you mention, but there was a
tutorial at CEC in 2005, and two tutorials at GECCO (intro and
advanced) last year ... and the same will be true this year.
Additionally, we started a terminology wiki after the discussion forum
several years ago, though it admittedly needs some work.  GECCO
routinely has about dozen or so papers on coevolution each year,
there's typically a FOGA and Dagstuhl presence for the community, as
well.  There was even a workshop in 2005 at the AAAI Fall symposium on
coevolution and coadaptation.

Still, as active as it is, it's still a very small group.  And of
those that USE coevolutionary algorithms, I think an even smaller
subset that STUDY coevolutionary computation ... and I've the
impression that the former pay little attention to the latter.  The
recent activity has produced some really great research (in my
opinion) and our foundational understanding of these systems have
greatly improved, but I don't get the sense that too many more people
in the wider field of EC are paying much attention to CoEC.  Moreover,
my feeling is that many of the issues with which CoEC deals are the
same in some other, non-EC, fields (e.g., multiagent reinforcement
learning) ... but these fields do not communicate much at all with
us.  When they do, the language and goals are so different, it is
challenging just to synchronize.

To that end, I wonder if the right role for this "task force" is to
construct a committee whose main goal is to actively PROMOTE
coevolution as a serious topic of study.  This might involve a special
issue in a major journal, but the special issue should have some kind
of goal beyond simply collecting similarly topic'ed papers:  Perhaps a
theme of how and why coevolution is useful and interesting to a wider
community should be strongly emphasized.  Moreover, perhaps we want
(additionally) to look outside the EC community (e.g., the new
"Natural Computing" journal, or something related to multiagent
learning and/or RL) .

Indeed, between us, we probably attend a wide variety of
professional / scientific events (WCCI 2008, as well as others).
Perhaps a concerted effort to bring contemporary CoEC research topics
to other fields by way of workshops, tutorials, discussion forums,
etc. might be useful.  With a little coordination, we might even be
able to reduce the workload somewhat by sharing what we have, etc.
But I think it's worth considering whether our goals should be more
specific than simply having a coevolution event.  I attend most of
these events, and they are great and personally quite rewarding ...
but I don't get the sense that they have really changed the state of
the field in any substantive way.

So perhaps we should think about our overarching goals:  What do we
want and/or expect from such events?  More researchers studying
coevolution?  More practitioners using coevolutionary algorithms?
More increased activity between machine learning fields?

For example:  As a theorist, one of my interests is in getting even
the relatively small community of people who use CEAs to consider
contemporary theory more than I believe they currently do.  I'm not
saying that should be the task force's goal ... but I do believe the
task force would do well to consider these high level questions.

(BTW:  One of my goals in writing this message was to finish before my
morning coffee got cold, and I've failed ... which means I've got
nasty coffee and the message is too long.)

I guess the big question is: to what task would our collective force
most intelligently be applied?  I don't know the answer to this
question, so perhaps we ought to start with brainstorming.  Hey, wait
a minute, "brainstorming" is my answer!  To that end:

I also do theory, and like Paul I also find that application papers
tend to be stuck in a 10-year-old (hence way outdated) mindset.  In
the broader EC community, coevolution is paradoxically recognized as
an important area of study while being treated as a footnote.  For
example, coevolution is briefly mentioned in Chapter 5 of Bill
Langdon's GP book and in Chapter 7 of Ken De Jong's EC book.  A recent
FOGA presentation on EC theory by Darrell Whitley completely neglected
coevolution, even though the theory of coevolutionary algorithms does
not appear to suffer from the pitfalls he was describing (and should
be applauded for that, right?)

Will promotion change any of this?  As Paul emphasized, we've made
some strong attempts at promoting recent developments, not to mention
the appearance of several influential PhD dissertations in the last
five years.  Will it take a Sims- or Hillis- caliber killer app paper
to grab people's attention?

Personally, I can envision ideas from coevolutionary algorithms
spilling into and informing questions in EC.  Some results in that
direction might be noticed (we might submit a coevolution paper to the
GA track at GECCO).  For instance:

* Adaptive representations: conceiving of genes as coevolving in the
"environment" of a genome, can we import ideas and insights from
coevolutionary algorithms research into questions of what
representation is/should be doing?  Richard Watson's SEAM algorithm
starts down this path but needs followup.  Ken's NEAT has a similar
flavor (at an abstract level, anyway).  Sevan and I recently observed
there is a connection between Ken's ideas of elaboration/
complexification and the kind of elaboration which occurs when using
mixtures (Nash) or sets (Pareto) of strategies.  There are questions
of cycles/overspecialization happening at the level of schema, for
example (I'd guess that the coevolutionary notion of
overspecialization can happen when you think of schema as coevolving,
and that it manifests in a GA as a local optimum).

* Niching/diversity maintenance: what do you want out of a niche (or
in it, for that matter)?  What kind of diversity should a population
have?  We see concepts in coevolution like Rosin's fitness sharing,
Juille's fitness sharing or,more recently, Cartlidge's parasite
virulence (Dan Ashlock had at least one paper on virulence too, but
this idea has a history back to Hillis, maybe earlier).  In most of
these works the diversity is *evaluational* or *performance* oriented,
rather than genotypic as in EC.  However, it's conceivable that
something like this is true: "useful" first-order genetic diversity is
equivalent to second-order performance diversity (anyone want to
precisely interpret that statement with me and write a paper about
it?)  Meaning, roughly, that the only genetic diversity which really
matters ought to confer some performance (read: fitness) benefit a few
generations down the line.  And if that's the case, ideas about
evaluational or performance diversity coming from coevolution might
have something to say.  I know Jeff Horn's recent shape nesting stuff
fits in here somewhere but I haven't  gone through the exercise of
nesting that shape into this picture.

* Learnable test sets:  Shiva had a nice paper at that AAAI Fall
Symposium on coevolution/coadaptation which I feel makes a bridge
between questions of evaluation on the one hand and questions of
representation on the other.  I'm running out of steam, but the basic
idea goes like so.  Start with an individual and consider all possible
offspring of it (Shiva considers only single-point mutations in that
paper).  You end up with a set of individuals which you can conceive
of as a population; then you can apply your favorite population-level
ideas to that set.  Shiva talks about forgetting (does a mutant of X
fail to do something which X could do?) but you could just as well
talk about diversity, elites, ...anything you'd want to say about a
population.  Notions from Pareto coevolution show up here: if X's
mutant dominates X, then the mutant has not "forgotten" anything, and
may have "learned" something.  The notions give a nascent declaration
of what representation "should be" doing

Anyway, I'm babbling.  My main point is that to get the attention of
another group, you need to solve a problem of interest to that group
using your own technique (where "problem" might be a theoretical or
conceptual problem).  I've personally tried and failed to do that with
machine learning people, but I think more modestly we can aim at EC/
GECCO and do some cool stuff.

More broadly, I guess I am arguing in favor of brainstorming with the
aim of building bridges to other communities.  We're too small a
community to get locked into rigid notions of what our field has been,
is, and should be.  We'll be easily ignored by the broader community
as we steep in our own juices.  Not to say that anyone present is
doing that, but that kind of ossification will happen on its own if we
don't actively resist it, in my opinion.  One thing we might do is
toss around as many ideas as we can about how coevolution applies to
other, related fields.  Which, now that I read it again, sounds like
exactly what Paul was saying.

Anyway, I've stormed my brain enough.  Best,

Anthony

On May 21, 9:24 am, "R. Paul Wiegand" <p...@tesseract.org> wrote:

Hence I like Paul and Anthony's talk along the lines of making CoEC
more accessible/visible to the EC community (and maybe wider). Perhaps
a special issue in "Natural Computing" as Paul suggests. Seems like
there are plenty of ideas (as Anthony illustrates) that would
reinforce the suggested theme. It would be good to hear other people's
ideas too. I wonder if it might be worthwhile to have some sort of get
together at CEC in Singapore (or before if there is a suitable venue -
I mention CEC because I will be there - maybe GECCO would work better
for others?) to continue brainstorming?

phi

On May 21, 11:27 pm, abucci <abu...@gmail.com> wrote:

To wit:  If you will be attending GECCO and you want to be involved in
a high-level goals-oriented discussion about the IEEE Taskforce on
Computational Intelligence and Coevolution, let me know (use "Reply to
author", rather than a "Reply").  If there's enough interest, I'll set
something informal up.

If the group is small enough, perhaps we can just meet for lunch or
dinner one day.

Thank you very much for volunteering to organise this! It would be great
if you could post up a report after GECCO. Wish I was going - it looks
like it's going to be an interesting conference this year.

cheers, phi

Associate Professor Philip Hingston
School of Computer and Information Science
Edith Cowan University
(+61 8) 9370 6427

CRICOS Institution Provider Code 00279B