385 International Journal on Advances in Intelligent Systems, vol 8 no 3&4, year 2015, http://www.iariajournals.org/intelligent_systems/ 2015, © Copyright by authors, Published under agreement with IARIA - www.iaria.org A Novel Chemistry-inspired Approach to Efficient Coordination of Multi-mission Networked Objects Mahmoud ElGammal The Bradley Department of Electrical and Computer Engineering Virginia Polytechnic Institute and State University Blacksburg, Virginia 24061 Email: gammal@vt.edu Mohamed Eltoweissy * Department of Computer and Information Sciences Virginia Military Institute Lexington, Virginia 24450 Email: eltoweissymy@vmi.edu Abstract—In this paper we present a chemistry-inspired approach for coordinating networked objects in pervasive computing environments built on the concept of chemical affinity. Our thesis is that by paralleling the model of interaction that takes place among atoms during a chemical reaction, a form of collective intelligence emerges among the objects in the network enabling them to achieve a common global objective while relying solely on preferences expressed on an individual basis. The main contribution of this paper is a novel implementation of a highly-parallelized chemical reaction execution engine that uses message passing to optimize reactant selection for multiple reaction rules simultaneously. In our method, objects in the chemical domain are represented using a probabilistic factor graph, where inter-reactant affinities are encoded in the factor nodes to guide bond formation among reactants. The problem of associating reactants with reaction rules is modeled as a Maximum-a-Posteriori (MAP) assignment problem, which we solve using the Max-Product Belief Propagation algorithm, allowing us to efficiently obtain a reactant-to-reaction assignment that maximizes the number of concurrent reactions. To evaluate our approach, we use simulation to assess the performance of the reaction execution engine in terms of execution speed and solution quality. Finally, we use the problem of resource-constrained task assignment among heterogeneous robots as a case study to present a concrete application of our approach. Keywords–Nature-inspired computing; Internet-of-Things; Pervasive computing; Belief propagation; Computational Chemistry. I. I NTRODUCTION With the current surge in smart computing applications, researches are increasingly turning to nature-inspired computing models for ideas on how to deal with the highly dynamic nature of this new computing paradigm. In [1] we proposed a new approach to network configuration and management in pervasive computing systems inspired by the chemical affinity concept, which we coined C 2 A 2 : Chemistry-inspired, Context-Aware, and Autonomic Management System for Networked Objects. The concept behind C 2 A 2 is that physical and logical components of the network are mapped to the chemical domain using a layered * Also Affiliate Professor, The Bradley Department of Electrical and Computer Engineering, Virginia Polytechnic Institute and State University, Blacksburg, Virginia, USA. abstraction model. Once represented using the chemical metaphor, reaction rules are then defined to specify how reactants in the chemical domain are allowed to interact with each other, which would eventually lead to implications on the actual network. C 2 A 2 relied on a reaction execution engine that was responsible for deciding which reaction rules may be fired and which reactants should be consumed by them. The work presented herein serves as a more in-depth discussion of a significantly improved implementation of the reaction execution engine previously introduced in [1]. In our new approach, reactants and reaction rules in the chemical domain are modeled using a probabilistic factor graph, where factor nodes encode affinities between each pair of reactants, affinities between reactants and reaction rules, as well as the different constraints needed to ensure that the resulting solution constitutes a valid reactant-to-reaction assignment. The graph is constructed such that the optimal assignment of reactants to reaction rules can be obtained by solving the Maximum-a-Posteriori assignment problem [2] on the graph, which we solve by passing carefully designed messages over the graph according to the Max-Product Belief Propagation algorithm [3]. A key advantage of this approach is its ability to find a solution that maximizes the number of reaction rules that can be satisfied simultaneously, which makes it particularly suited for multi-mission pervasive computing applications. The remainder of the document is organized as follows. In Section II, we survey some notable related works from the literature and provide an overview of the probabilistic graphical modeling techniques we rely on in later sections. In Section III, we present the implementation of our reaction execution engine. In Section IV, we analyze the performance of our approach using simulation, and validate its efficacy by applying it to a more concrete case study. Finally, in Section V, we present our conclusion and discuss future work. II. BACKGROUND AND RELATED WORK In C 2 A 2 we leverage ideas from the fields of bio-inspired computing and machine learning. In particular, we build our work upon some of the established concepts and algorithms belonging to the subfields of chemistry-inspired