Tuesday, July 19, 2011

Self-organization is "supra-optimal"

Thanks to a post shared by Eric Le Goff on Google+, I ran across a very interesting article describing some fundamental research on neuronal network organization.  The authors have developed a research "platform" that enables them to grow physical neuronal networks on chips.  They are using the platform to observe self-organizing behavior of the networks.  What is becoming clear from this and other research is that the large-scale architecture and macro behavior of our brains is most likely the result of self-organizing collaboration among a hierarchy of network clusters, each "operating" with locally conditioned architecture and control.

After reading the article, I was embarrassed to notice that I had not yet heard of Plos One, where it appeared.  Poking about Plos One, I stumbled on yet another article on self-organizing behavior.  Somebody must be trying to tell me something.   The second article is a nice visceral illustration of how we need to think differently about how to manage complex systems.   The authors look at train scheduling in public transportation systems.  They show that "global" scheduling systems for train arrival and departure times can be beaten by a system that imitates ant colony behavior - trains emitting "anti-pheromones" that signal trains behind them to modify their behavior.   What is interesting here is that independent decision-making based on locally available information can result in better global outcomes than a globally engineered solution.

These ideas are not new.  What these two articles illustrate, however, is how much change is going to be required in how we think about modelling and engineering to take advantage of them.  For example, the whole concept of "control" in the pure engineering sense is going to have to be transformed to engineer the kinds of systems that the second article hints at.  Imagine trying to apply six sigma to that transportation system.   Similarly, the first article indicates that it is likely hopeless to try to get much more than we already have from macro-level decomposition of the brain or functional/reductionist analysis of isolated neuronal networks.

I also notice that Plos.org has opened up their search APIs and (jointly with Mendeley) launched "a contest to build the best apps that make science more open."  Hmm...somebody is really trying to tell me something here.

No comments:

Post a Comment