"Machine Learning" And "Frontiers In Evolutionary Computation"

I have read a post from our blog friend Julian Togelius concerning key scientific challenges in machine learning. It is from a researcher at Yahoo! It's funny: When someone tries to set up a list of top something (facts, songs, challenges, whatever...) the bias towards the author's own area of interest seems almost inescapable.

It is an extremely limited list of "challenges" and I would not say they are challenges at all. Further I explain why and point to a similar situation I once had while reading an edited evolutionary computation "book". But before begining my arguments, let me tell you that I do not intend, in any way and in any sense, to slur the work/research of anyone else out there. It is just my own opinion and not the real true itself.

Once I read some pages of an evolutionary computation book named "Frontiers Of Evolutionary Computation", an edited volume in which there were some well known researchers stating what they see as a frontier in that field. Despite the grandeur book title when we take a look at the crude reality we realize the most known and applied evolutionary algorithm still is the good old elitist Simple Genetic Algorithm (SGA), which dates back to the early 1970s (or late 1960s). In my opinion, that could be the main frontier in evolutionary computation today: Why, despite all the new types of evolutionary-based algorithms, the most known and used still is the good old elitist SGA? That is ironic. I think the main reason for the elitist SGA's big mainaasuccess is due to the heuristic knowledge its users have embedded in it.

Maybe, the book was not so much about frontiers in evolutionary computation, but research problems the authors were facing and those problem may or may not represent a frontier in that research area -- therefore, I consider a more honest book title would be "Guess What??? We Are Still Using The Good Old Elitist SGA". The same is valid for the Yahoo! guy: I consider that list he made was not composed of key scientific challenges in machine learning, but only key information technology problems Yahoo! has faced. Those problems can be solved through the knowldge science has to give us.

Just a final word about the aforementioned edited book. It seems our time is living an interesting, let's say, post-modern times phenomenon: Edited books. Springer has lots and lots of them, ranging from some well obscure book titles and areas to subjects that hardly will find a passionate reader -- surely, there are nice titles too. Nowadays, anything is eligible to become an edited book: From umbrellas to telephone cabins. I look at those books with a grain of salt: I doubt if, indeed, there is a nice amount of interest on them. For example, take a look at how many persons have bought the book above at Amazon.com.

I hope that some well regarded journals do not endeavour in such a practice. Otherwise, soon we will see journals like "IEEE Transactions On Telephone Cabins".