Kostas Stergiou has published in Artificial Intelligence Review a very interesting “review and evaluation” of adaptive constraint propagation in constraint satisfaction. It surveys, classifies, and advances the subject, and suggests directions for further work.

With the current prominence of machine learning, I was happy to see that Stergiou unearthed an old paper Rick Wallace and I wrote on Selective Relaxation for Constraint Satisfaction Problems, which might be regarded as including an early application of machine learning to CP. That presupposes a rather generous definition of “learning”, which might even apply to another of the cited papers, one I wrote with Dan Sabin on Understanding and Improving the MAC Algorithm. “Adaptation” might be a better term. A more straightforward example of learning is the paper cited that I wrote with Susan Epstein, Rick Wallace and Xingjian Li on Learning Propagation Policies. Kostas concludes his proposals for future work with a timely call for exploring the further potential of machine learning. (Incidentally, I ran across a paper on ArXiv recently that applys deep learning to the related issue of variable ordering.)

I’d like to see more survey papers being published. It occurs to me that every thesis will include the makings of such a paper in a “related work” section or chapter. I would encourage students not just to publish papers based on the novel results in their thesis, but to publish a survey article based on their thesis work as well. The Constraints journal even has a “Surveys Editor”, currently Pascal Van Hentenryck.

I have added a link to Kostas’ paper in the “Surveys/Overviews/Bibliographies” section of my Resources website. If you publish a new survey, or have published one that is not already linked to at the website, let me know so I can link to it.