, , , , , ,

If you don’t know anything about the world – and if your life depends on it – you have to use trial and error to find out what is going on in your environment.  In this situation it’s the only way you can learn, and it’s the basic manner in which evolution discovers survival strategies in populations. Evolution consists of an iterative cycle of reproduction with variation, environmental selection of the best strategies, followed by new rounds of reproduction, and so on.

Exploring this approach to solve difficult practical problems is the focus of the community behind the Genetic and Evolutionary Computation Conferences and also the GECCO 2011, which July 13-16, 2011, gathered some 600 delegates in Dublin. They are scientists and engineers as well as practitioners from industry, business, policy and the arts. This community meets every other year at different locations to discuss results, problems and visions for utilizing computational methods based on a variety problem solving strategies inspired from biological systems.

I can’t do justice to the many interesting presentations and ideas showcased at this meeting, so please have a look at the program here. For the non-scientists, perhaps the many GECCO competitions are the most intriguing and accessible. They include the evolution of race-car autopilots, development of financial time series predictors, the evolution of the best sounding music as well as websites where kids can evolve intriguing 3D objects, have them printed, and then sent them to their home addresses.

Since I’m not an expert on evolutionary computational methods and thus a bit of an outsider to the GECCO community, I’ll share a few observations below, which I also discussed with old and new colleagues during the conference.

In the 20 years the GECCO community has existed, it has indisputably proven the validity of using evolutionary approaches. These methods can indeed solve hard problems in many areas. That is a significant achievement and an advancement of science. My guess is that this success (the methods really work!), combined with a fascination with evolution, are the main reasons for the vitality and the continued growth of the community.

However, the community, as most new professional communities, has a bit of a tribal feel. With all its successes, I suspect the GECCO community could benefit from more interactions with the traditional operational research (OR) and statistics modeling communities. It would be good to compare and contrast these evolutionary and swarm based approaches with more classical OR and statistics problem solving methods.  Once you learn something about your problem / environment you can do better than the naïve trial and error searches.

To use a biological metaphor: Life evolved successfully for some 3+ billion years solving survival problems only utilizing the naive trial and error method. However, at some point life invented intelligence, which turned out to be an extremely efficient survival skill.

In this context, intelligence means that individuals can learn about their environment, or their “fitness function”, based on their previous experiences. This means that an individual doesn’t always need to risk its life on simple trial and error learning, but can react and learn more about its environment based on what it already has learned.  For example, if you once walked close to a big striped cat and were almost eaten, you should probably run before you get close to a big black cat. You don’t need to repeat this “getting close to” experiment when you see another big cat even if the cat has a different color.

For the computing analogy this means that as you learn something about your environment, your fitness function, you can utilize this knowledge to develop a more efficient search strategy for finding solutions to your problem. You are no longer dependent on an unbiased trial and error process. Most real-life problems have some structure to them (correlations in their fitness functions), which means that intelligent searches come out ahead of naive evolutionary- or swarm searches.  Also, many real life problems involve co-evolving individuals, which together change the environment and thus the involved fitness functions, which is a further challenge for problem solving.

A continued exploration of what evolution can do for you as a problem solver, as well as addressing some of the mentioned above issues, is in part where I believe the GECCO community is heading.

Besides the many lectures and technical discussions, I certainly enjoyed the Irish hospitality, the lively pubs, the visit to the Guinness Brewing Company, the Irish folk dance, and a swim in the beautiful Irish Sea at Howth.  I must admit the latter was a bit cold even for a seasoned Dane.