J.P. PABICO. 2008. (INVITED PAPER) First DOST-PCASTRD National Symposium on Science and Technology (NSST 2008), SEARCA, UPLB, 08-09 October 2008.

Abstract

In this paper, we present a distributed stochastic algorithm that simultaneously solves several real-world problems by simulating the interaction dynamics of abstract entities. The abstract entities encode solutions to the problems. The algorithm is inspired by molecular dynamics such that the objects are abstract entities or data and the interaction among them is driven by an algorithm. Here, the object has a dual characteristic, acting either as a machine (operator) or as a data (operand), and thus it can either process other objects or be processed. We exploited this dualism of objects to implicitly define a constructive computational procedure using the chemical dynamics as a metaphor to solve real-world problems. We will illustrate the mechanism of this computational procedure by presenting the respective mappings of the solution spaces of search, optimization and machine learning problems. We will use as test beds the traveling salesperson problem, the computation of cultivar coefficients in CERES-Rice plant growth simulation model, and the grammar inference given positive and negative examples of a language for search, optimization and machine learning, respectively. The artificial object that exists in an artificial chemical reactor encodes a Hamiltonian cycle, a set of coefficients, or a grammar. The artificial chemical reactor is governed by interaction rules that can re-order a Hamiltonian cycle, alter coefficient values, or re-rule a grammar. The cost of the Hamiltonian cycle, the mean square error between the simulated rice growth and the actual observed rice growth data, and the total number of accepted positive language examples and rejected negative language examples are considered as the molecular mass. This metaphor, tested by solving in tandem with a deterministic algorithm, has been shown to find quality solutions in finding the minimum Hamiltonian cycle cost, in optimizing the set of CERES-Rice cultivar coefficients, and in learning a grammar.

Keywords: artificial chemistry, computational metaphor, search, optimization, machine learning.