Didn’t I just say “Woo hoo” yesterday? False alarm. Scarcely do I clear one set of major tasks away than another set rise up. I already mentioned that I was going to be the speaker at the Humanists of Minnesota banquet on Saturday evening. I neglected to tell you all that I’m leaving for the University of Michigan tomorrow to give the keynote at the Genetic Programming Theory and Practice Workshop.
I know virtually nothing about genetic programming, so this is a wonderful opportunity to learn something about it.
Since I’m certainly not going to be able to tell them a thing about genetic programming, I’m planning to tell them a little about my own skewed perspective as one of those metazoan-centric fans of developmental processes. I’m hoping they might learn a little something from me, and that we’ll all have some fun with ideas about embryos. Here’s my very brief abstract:
A developmental biologist’s view of evolution
The ongoing integration of molecular genetics, developmental biology, and evolution (the field of evo-devo) is stirring up new ideas and new questions. I will tell a few stories from the evo-devo literature that illustrate the importance of the principles of developmental plasticity and developmental constraint on evolutionary trajectories — showing that these are two competing and complementary forces operating on multicellular organisms. My argument is that the contingencies of developmental architectures may well be as significant a force on evolutionary histories as selection.
Next week I get to slack off. No, wait, there’s also…
A major advantage to genetic algorithms and genetic programming, and a significant difference from the evolution of life, is that they can be and often are run multiple times, to see the different kinds of solutions that emerge. That doesn’t eliminate path dependency that went into the development of any particular solution. It does allow the modeler to see the different kinds of trajectories that converge on the specific problem.
Obviously, we can’t do that with the biological world. ;-)
Genetic algorithms are Bayes! EVERYTHING IS BAYES!
…
or so I’m told.
GA are also useful for non-linear optimzation problems, where gradient methods tend to barf.
Off topic:
Jerry Falwell just died.
I’d love to hear the talks at that workshop, although I’ve only just begun exploring genetic algorithms in my evorunners game so most of it would probably fly over my head!
Re:King Aardvark
No doubt a planned attack by abortionists, gays, Darwinists and the ACLU.
I’m bemused that this group of mostly computer scientists chose you as keynote speaker. Perhaps it’s that your specialty of developmental biology will address GP’s weakest point: few genetic programming projects make any distinction between ‘genotype’ and phenotype. In the early ’90’s Frederic Gruau’s dissertation and a few papers included a developmental process – but not much since. It’s not at all clear how pertinent ontogeny can be to GP dynamics.
The wikipedia article on GP links to their mailing list on Yahoo, and that list has a recent thread initiated by Douglas Mota Dias’s request for refs for his hour-long talk on GP variations and applications – the thread responses include the core historical papers from Turing to Koza.
My argument is that the contingencies of developmental architectures may well be as significant a force on evolutionary histories as selection.
Now I really want to see your talk. Are not developmental architectures themselves under selection?
May I suggest the Arbor Brewing Company for an evenings repast while you are in Ann Arbor. The pub has the most excellent brew. You are also welcome anytime at the N3rd household, but it is featuring late-night-class-for -Papa frozen fishstick special.
Your talk on evolution from a developmental biologist’s view sounds very interesting. Wish I could be there!
Perhaps you could post some examples that would prove your argument that the contingencies of developmental architectures may be as important as selection in evolutionary histories.
Cyan, I have also found that comment and its observation useful – it is certainly on topic here. The similarity between genomes learning (adapting allele frequencies) and general learning (adapting behavior) becomes striking IMHO.
(Even though I insist that bayesian models should be tested. TEST RULEZ! … or so I’m told. I guess testing of populations or behaviors is what happens, and is measured, by the time they go extinct. ;-)
Btw, you seem to have missed the context of the commenter’s name. That was on The n-Category Café and there it seems like EVERYTHING IS BAEZ! :-)
Cyan, I have also found that comment and its observation useful – it is certainly on topic here. The similarity between genomes learning (adapting allele frequencies) and general learning (adapting behavior) becomes striking IMHO.
(Even though I insist that bayesian models should be tested. TEST RULEZ! … or so I’m told. I guess testing of populations or behaviors is what happens, and is measured, by the time they go extinct. ;-)
Btw, you seem to have missed the context of the commenter’s name. That was on The n-Category Café and there it seems like EVERYTHING IS BAEZ! :-)
Oh, and this:
Or perhaps we could see a link to the OH’s later?
Oh, and this:
Or perhaps we could see a link to the OH’s later?
Torbjörn Larsson, OM:
I keep constructing puns on that, “Baezian prior” and so forth. For example, “Baezian learning” is what happens when you read This Week’s Finds in Mathematical Physics too late at night.
A friend of mine read TWF for six consecutive hours one evening. When he finally tumbled into bed, he dreamed he was arguing with a fellow physicist about some combinatorical problem in statistical mechanics. The two discussed the problem back and forth, until finally they agreed, “It all depends upon the cobordism between this and that.”
My friend then woke up screaming, “I don’t even know what the fuck a cobordism is!”
I built a whole programming language (http://www.transmuter.org) designed to explore this idea (which is temporarily on hold, due to survival-related issues). Hopefully I can start working on it again soon.
JeffW: Looks interesting (from first glance only). Your journal page is worth highlighting for its discussion of the biological interpretations.
Cyan: your link is interesting. The analogy (perhaps more) between learning and evolutionary dynamics is familiar to biologists and psychologists, of course, tho rarely so quantified. For example, this qualitative overview from 1993: Darwin Machines and the Nature of Knowledge, by psychologist Henry Plotkin. From its closing paragraphs:
JeffW: Looks interesting (from first glance only). Your journal page is worth highlighting for its discussion of the biological interpretations.
Cyan: your link is interesting. The analogy (perhaps more) between learning and evolutionary dynamics is familiar to biologists and psychologists, of course, tho rarely so quantified. For example, this qualitative overview from 1993: Darwin Machines and the Nature of Knowledge, by psychologist Henry Plotkin. From its closing paragraphs:
LOL!
And, poor thing, nerdmares can be rough.
LOL!
And, poor thing, nerdmares can be rough.