Bio-inspired Training Algorithms for Artificial Hydrocarbon Networks: A Comparative Study

Abstract

Artificial hydrocarbon networks (AHN) is a supervised learning algorithm inspired on chemical organic compounds. Its first implementation occupied the well-known least squares estimates (LSE) as part of the training algorithm. Unsurprisingly, AHN cannot converge to suitable solutions when dealing with high dimensional data, falling into the curse of dimensionality. In that sense, this paper proposes two hybrid training algorithms for AHN using bio-inspired algorithms, i.e. Simulated annealing and particle swarm optimization, and compares them against the LSE-based method. Experimental results show that these bio-inspired algorithms improve the performance of artificial hydrocarbon networks, concluding that these hybrid algorithms can be used as alternative learning algorithms for high dimensional data.