First International Conference Modelling and Development of Intelligent Systems Sibiu - Romania, 22-25 October, 2009 Wasp based algorithms and applications Dana Simian Abstract The aim of this paper is to present the wasp based computational model and many applications of wasp based algorithms. A general frame for designing a wasp based algorithm, starting from the classical problem of task allocation in a factory, is realized. The most important characteristics of the wasp computational model are underlined and the way of particularization of these characteristics for each problem is presented. Original applications of wasp based algorithms in modeling multi agent systems, in solving optimization problems and in building a reinforcement scheme for a stochastic learning system are presented. 1 Introduction In the last ten years methods and models inspired from the behavior of social insects like ants and wasps have gained increasing attention. Computational analogies to adaptive natural multi- agent systems have served as inspiration for multi-agent optimization and control algorithms in a variety of domains and contexts. Self-organization, direct and indirect interactions between individuals are important characteristics of these natural multi-agent systems. Metaheuristics inspired from nature represent an important approach to solve NP-difficult problems. It is important to identify when a problem can be solved using these kind of methods. It is the goal of this article to identify some type of problems which can be solved using wasp computational based algorithms and to give a general frame for design these algorithms. The remainder of this paper is organized as follows: in section 2 we present the wasp computational model and the classical problem of task allocation in a factory. Starting from this problem we realize a frame for design models based on wasp behavior and present many models for multi agents systems, from different fields. In section 3 we present a reinforcement scheme for stochastic learning automata, based on wasp behavior. In section 4 we present a wasp based algorithm for improving the performances of a co-mutation operator. The co-mutation operator is used in a hybrid approach for building multiple SVM kernels. Conclusions are presented in section 5. 2 Wasp behavior based algorithms in modeling multiagent systems The self organization model that takes place within a colony of wasps was used for solving large complex problems, most of them with a dynamic character. In [18], Theraulaz et al. present 1 A Constraint-Based Approach to the Timetabling Problem Cristian Fr˘ asinaru Abstract It is well known that timetabling problems are usually hard to solve and require a lot of compu- tational effort. There are many theoretical models that address this type of problems and various algorithms have been developed in order to attempt to solve them efficiently. However, it is not easy at all to apply these models to real life situations. This paper presents a solution to create an universal constraint-based model for representing the timetabling problem that can be applied in universities, schools or any other similar domain. Once the model is created it can be effectively solved with any CSP solver. We have used our own CSP solver, called OmniCS 1 (Omni Constraint Solver), that allows an incremental, human-aided approach to the timetabling problem, which proved very useful in practical applications. 1 Introduction A timetabling problem can be defined as the scheduling of some activities during a certain period of time. Each activity has a set of properties, like the participants who attend it or the resources it requires, and it is subject to certain restrictions regarding its possible planning. University timetables for instance must manage entities like courses, students, teachers and rooms in order to create a mapping between courses and the time-slots of the week. Usually, timetables cycle every week or every fortnight but this will not become a requisition of our model. Traditionally, the problem is solved manually and it is a tedious job thatrequires days or even weeks. Automated building of the timetables is also very difficult because there are many types of restrictions that must be accounted for and it is not easy to express them in computational forms. The timetabling problem has been studied intensively since the sixties ([11]) and different techniques for solving combinatorial problems have been used, such as graph coloring ([3]), integer programming ([13]), simulated annealing ([1]), tabu-search ([5]) or genetic algorithms ([2]). A survey can be found in ([16]). Despite the fact that these methods have given good results, using them in real life applications was not easy and quite counterintuitive not only because the complexity of the restrictions could not be formalized properly but also because solving algorithms could not be modulated to follow human judgement, an aspect which is very important in the interactive creation of the timetble. In recent years, many computationally difficult problems from areas like planning and scheduling have been proven to be easily modelled as constraint satisfaction problems (CSP) ([6], [18]) and a new pro- gramming paradigm emerged in the form of constraint programming, providing the opportunity of having declarative descriptions of CSP instances and also obtaining their solutions in reasonable computational time. As a result, constraint satisfaction techniques have been applied to the timetabling problem ([10], [15]).Because constraint programming received very much attention also from the industry, a lot of CSP 1 Omnics is freely available at http://omnics.sourceforge.net 96