DeepBio goes online!

Welcome to DeepBio website. This site is devoted to provide the lastest news regarding the DeepBio project, whose aim is the study bioinspired algorithms endowed of a deep architecture and their use to solve massively complex problems, as well as their deployment on such massively complex environments. To this end, we will place strong emphasis on ephemeral computing, self-* properties and complex systems. Please, check the About page for more details about the goals of the project and the Groups page for details on the members of the team. The site is organized as a blog, and therefore the visitor can not only check the existing content but also interact with us by posting comments.

Thanks for visiting us. We hope you find this site interesting, and become a regular visitor.


Review of CAEPIA 2018

caepia2018The CAEPIA (Spanish acronym for “Conferencia de la Asociación Española para la Inteligencia Artificial“) 2018 conference took place in Granada and it has been the eighteenth edition. It’s a biennial forum open to researchers from around the world to present and discuss the latest advances in artificial technology in Artificial Intelligence. The Doctoral Consortium is organized within CAEPIA, a forum for doctoral students to interact with other groups through the discussion of their thesis project and it has been also held an app development competition based on AI techniques. Also within the CAEPIA, several federated congresses and workshops are held. A brief resume of the conference can be seen in the next video.

All events took place in the palace of congresses of Granada, located in the center of Granada, near of the Genil river. We could attend MAEB, CoSeCiVi and other conferences, as well as workshops from DeepL, IndustriA 4.0., and others. We had the opportunity to enjoy lectures from distinguished invited speakers such as Sergio Guadarrama (from Google research) and Pedro Larrañaga who was recipient of the AEPIA2018 award for his professional career and his contribution to Bayesian Networks. The social dinner of the conference took place in the beautiful Hotel Abades Nevada Palace and we could taste the delicious gastronomy of Granada.

The scientific quality of the event was impressive as well. About a hundred researchers presented their latest work in a relaxed environment. The program was organized around 10 parallel tracks, comprising both general and invited sessions. DeepBio presented a work in MAEB about the application of genetic algorithms and deep learning in the videogame design. Our presentation [1] is included below.

Overall, the conference was brilliantly organized and we are now looking forward to the next edition. Hope to see you there!

[1] A. Gutiérrez Rodríguez, C. Cotta, A.J. Fernández Leiva, An Evolutionary Approach to Metroidvania Videogame Design, XVIII Conferencia de la Asociación Española para la Inteligencia Artificial, pp. 518-523, F. Herrera et al. (eds.), Granada (Spain), 2018

New paper on the placement of suicide bomber detectors using metaheuristics

Suicide bombing is one of the prevalent ways in which terrorist organizations substantiate their infamous actions. The defense from this kind of attacks is a highly cross-disciplinary endeavor that can be approached from many points of views. From a tactical point of view, it is often useful to analyze it from a political-rational perspective, that is, using a cost-benefit analysis analogous to the rational calculations of a military staff or a corporate board.

Some works in the literature have taken some steps in this direction. We have built on our previous research in which we focused on the problem of determining the distribution of detectors on a given area under threat, aiming to maximize the probability of detecting and neutralizing suicide bombers. This problem, seminally approached by greedy techniques in the literature, was tackled with metaheuristics in our work

As in the previous literature, the simplistic assumption that the terrorist attacker would pick a certain target and the path to reach it uniformly at random was used. We have taken a step beyond this premise, and considered the problem from two perspectives. In the first one, the terrorist attacker takes weighted decisions based on the relative importance of each target. In the second one, we take a game-theory perspective: two parties are involved, the attacker and the defender; the defender tries to place detectors so as to minimize the inflicted damage received by the attacker; conversely the attacker tries to select the target and the path to reach it so as to maximize this inflicted damage, knowing that the defender will try to negate these attempts. From the point of view of the defender that has to place the detectors, this amounts to minimize losses in a worst-case scenario.

Springer's Natural ComputingWe have approached the resulting problem in a recent paper entitled

considering hill climbing, tabu search and evolutionary algorithms. The evolutionary algorithms reveals itself as particularly well-adapted for this problem when the worst-case perspective (which provides a more challenging optimization scenario) is used. The full abstract of the paper follows:

We consider an operational model of suicide bombing attacks –an increasingly prevalent form of terrorism– against specific targets, and the use of protective countermeasures based on the deployment of detectors over the area under threat. These detectors have to be carefully located in order to minimize the expected  number of casualties or the economic damage suffered, resulting in a hard  optimization problem for which different metaheuristics have been proposed.  Rather than assuming random decisions by the attacker, the problem is approached by considering different models of the latter, whereby he takes informed decisions on which objective must be targeted and through which path it has to be reached based on knowledge on the importance or value of the objectives or on the defensive strategy of the defender (a scenario that can be regarded as an adversarial game). We consider four different algorithms, namely a greedy heuristic, a hill climber, tabu search and an evolutionary algorithm, and study their performance on a broad collection of problem instances trying to resemble different realistic settings such as a coastal area, a modern urban area, and the historic core of an old town. It is shown that the adversarial scenario is harder for all techniques, and that the evolutionary algorithm seems to adapt better to the complexity of the resulting search landscape.