Yesterday I presented my work Migration Study on a Pareto-based Island Model for MOACOs, accepted as full-paper at the Genetic and Evolutionary Computation Conference 2013, held in Amsterdam.
The paper abstract is:
Pareto-based island model is a multi-colony distribution scheme recently presented for the resolution, by means of ant colony optimization algorithms, of bi-criteria problems. It yielded very promising results, but the model was implemented considering a unique Pareto-front-shaped unidirectional neighborhood migration topology, and a constant migration rate. In the present work two additional neighborhood topology schemes, and four different migration rates have been tested, considering the algorithm which obtained the best results in average in the model presentation article: MOACS (Multi-Objective Ant Colony System). Several experiments have been conducted, including statistical tests for better support the study. High values for the migration rate and the use of a bidirectional neighborhood migration topology yields the best results.
Yes, the title is a representation of how long the work is :D
The abstract is:
Multi-objective algorithms are aimed to obtain a set of solutions, called Pareto set, covering the whole Pareto front, i.e. the representation of the optimal set of solutions. To this end, the algorithms should yield a wide amount of near-optimal solutions with a good diversity or spread along this front. This work presents a study on different coarse-grained distribution schemes dealing with Multi-Objective Ant Colony Optimization Algorithms (MOACOs). Two of them are a variation of independent multi-colony structures, respectively having a fixed number of ants in every subset or distributing the whole amount of ants into small sub-colonies. We introduce in this paper a third method: an island-based model where the colonies communicate by migrating ants, following a neighbourhood topology which fits to the search space. All the methods are aimed to cover the whole PF, thus each sub-colony or island tries to search for solutions in a limited area, complemented by the rest of colonies, in order to obtain a more diverse high-quality set of solutions. The models have been tested by implementing them considering three different MOACOs: two well-known and CHAC, an algorithm previously proposed by the authors. Three different instances of the bi-Criteria travelling salesman problem have been considered. The experiments have been performed in a parallel environment (inside a cluster platform), in order to get a time improvement. Moreover, the system scaling factor with respect to the number of processors will be also analysed. The results show that the proposed Pareto-island model and its novel neighbourhood topology performs better than the other models, yielding a more diverse and more optimized set of solutions. Moreover, from the algorithmic point of view, the proposed algorithm, named CHAC, yields the best results on average.
The scheme of the proposed model can be seen in the next figure:
The last Wednesday (13 of May), we presented (again) our Multiobjective Ant Colony Optimization algorithm (yes, the famous CHAC :D) at NICSO 2010, which was held in Granada, in the same building where we work everyday…
… what a so far trip… :-| :D
The paper presents a study of the objective balancing parameter (named LAMBDA), used in this algorithm. ;)
I returned from Brussels a couple of days ago, where I went to present KANTS: Artificial Ant System for Classification (the model was already described here) at the 6th International Conference on Ant Colony Optimization and Swarm Intelligence. ANTS 2008 is similar to PPSN, with most of the papers being presented at poster sessions (only a few are chosen for oral presentation). This procedure works well when the poster sessions are not just a minor event of the congress, thrown to a distant room in the hotel/university where nobody even bothers to go, or scheduled to the end of a long day. ANTS sessions were well organized and every poster had an assigned space. My presentation was scheduled to the last day of the congress, when most of people had already packed for their trip back home, but nonetheless the session went well, with lots of people wandering around the room, clearly interested in the works. KANTS got the attention of some audience, and I think they were quite impressed by the simplicity (and efficiency) of the idea. The inevitable question arouse (are you planning to test KANTS on a real-world problem?) and this time we can answer yes, we are not only planning to do it, but we are already working on it (later we will report on those experiments). In the same section, another swarm-clustering was presented. I saw the poster, and the results on clustering appear to be quite good (but the system does not perform classification). I haven’t read the paper (as a matter of fact, it is published as an extended abstract), but I was able to realize that the algorithm is little bit complex, simulating the behavior of three different species: ants, birds and spiders.
A week before I was in Barcelona, at the 8th International Conference on Hybrid Intelligent Systems (HIS 2008), presenting the paper Tracking Extrema in Dynamic Fitness Functions with Dissortative Mating Genetic Algorithms. It is quite a different work, and more related to my thesis’ subject, bio-inspired computation on dynamic environments. It describes the experiments performed with an adaptive dissortative mating GA (ADMGA) on Dynamic Optimization Problems. Dissortative mating appears frequently in nature and refers to the occurrence of mating between dissimilar individuals more often than expected by chance. It maintains genetic diversity at a higher level, thus increasing the exploration stage of the algorithm. Dynamic fitness functions are more sensible to genetic diversity than static ones, and so dissortative mating is a good candidate to deal with that kind of problems. The paper describes mainly the experiments performed with trap functions and show that ADMGA may improve GAs on some dynamic optimization scenarios. Robustness is also addressed and results show that ADMGA maintains a more stable performance over the wide range of dynamic scenarios. The congress HIS is mainly dedicated to hybrid models and real-world applications, so ADMGA was somewhat “lost” among other works. But good news came after the congress, and this line of research will probably make its way through other media.
P.S. In Brussels, avoid Hotel Continental, near Midi Station, unless you need inspiration for another insects’ heuristics.
Next week, Carlos Fernandes will present in the ANTS 2008 conference our paper KANTS: Artificial Ant System for Classification (hope the typo is not in the proceedings, but I’m afraid it will be). The algorithm was already presented by Antonio in ALIFE XI, with the paper KohonAnts: a self-organizing ant algorithm for clustering and pattern classification (which is also available from arxiv). Antonio was questioned about what was good about this algorithm, and I guess this is as good a place as any other to tell about it.
The basic idea of Kohonants is to use stigmergy for clustering and classification. Usual ant clustering algorithm place data as objects in the grid ants move around, and then, via some natural inspiration and a great deal of heuristics, they manage to cluster them according to proximity.
Kohonants, on the other hand, makes each data item an ant (or several, if needed). Pheromones are also vectorial in nature, in the same dimension as data, and what ants do when they move about is first take into account what’s the pheromone levels they have around in their neighborhood, and second modify it making them closer to the vector they represent.
That is why they are called Kohonen’s Ants: Kohonen’s algorithm behaves in the same way. Takes a training data vector, compares it to all the vectors in a two-dimensional array, and whoever wins is made closer to the data vector. Ants in Kohonants take the place of data vector in Kohonen’s algorithm, and the two-dimensional vector array that is trained is substituted by the two-dimensional (vectorial) pheromone field in Kohonants.
Results so far have been quite good, but we’ll continue with it to see what are their limits, and how well it fares against other ant and non-ant clustering algorithms. Meanwhile, as we mentioned in our previous post, you can download full code from the GeNeura code repository