My Constraint Programming Blog

Wednesday, Day 2

Today started and ended with two very inspiring talks, both showing how to use optimization for important issues, and both also showed the need for multi-disciplicary approaches for these kind of projects, i.e. not only CP or optimization in general but also machine learning, simulation, agent-based modeling (all areas in which I've been interested in on and off, especially machine learning/data mining).

Michaela Milano started the day with the invited talk Optimization for Policy Making: the Cornerstone for an Integrated Approach where the general goal is to build a policy making system (decision support) for energy consumption.

The last lecture was Pascal Van Hentenryck's exiting public lecture Decide Different! about many projects he and his collegues at NICTA (Australia) and other universities was involved in, such as disaster recovery. He also showed two interesting example of optimization in crowd funding and similar systems.

A related aside: Pascal is also behind the very interesting Coursera online Course Discrete Optimization. I have not yet seen all the lectures, but the one about Dynamic Programming and Constraint Programming was excellent, especially for introducing CP. (I will later see the rest of the course, Local search and Integer programming, and perhap be more serious and do some of the homeworks). [In his wonderful introduction lecture, Pascal showed Kidney Exchange as an example of a hard discrete optimization, and that inspired me to write a small MiniZinc model to solve these kind of problems, though it cannot at all handle the huge number of people that is required: about 80000 people is of a need of a kidney. My model solves a problem with 500 people (randomly generated) fairly simple, and 1000 with some more sweat.]

Some other interesting talks today: In the Search track Geoffrey Chu talked about "Dominance Driven Search" (yet another approach that might lead us to a powerful black box solving?), and Marc Shoenauer talked about Bandit-based Search for CP (which - to be honest - was more interesting than I first anticipated). Jean-Charles Régin's talk about Embarrassingly Parallel Search showed a way which seems to be a good way to do massive parallel search (though I'm definitely not an expert in this area).

Ian Gent and Lars Kotthoff held a kind of iconoclastic tutorial about recomputation, which may very well be the future way of doing repeatable experiments in computer science (and surely in other disciplines that use computer experiments). It seems to be quite easy both to submit an experiment to the site and especially to install and redo the experiment. See http://www.recomputation.org/ for more information. The tag line for the blog/project is If we can compute your experiment now, anyone can recompute it 20 years from now. Six of the accepted papers has their experiment at Recomputation.org, see here. For a simple demo experiment, see Our First Experiment: A Chess Puzzle.

My favorite talk from the Global Constraint track was "Solving String Constraints: The Case for Constraint Programming", where Pierre Flener presented the paper by Jun He et.al. The paper compared a CP approach with dedicated solver such as Hampi, Kaluza and Sushi for certain string constraint benchmark. The CP solver dominated these solvers completely. (I actually played with Hampi this summer missed much of the things I'm used to with a CP solver.)

Tomorrow (Thursday) will be interesting. It starts with a presentation of a new CP system: Objective-CP (a CP solver written in Objective-C) and ends with the First International Lightning Model and Solve Competition . I'm also looking forward to Tias Guns' ACP Doctoral Research Award: "Declarative Pattern Mining using Constraint Programming". Oh, sorry. The day actually ends with the Conference Banquet.

(Greeting list today: Barry Hurley, Geoffrey Chu, Laurent Michel.)