摘要
The conventional method for parameter estimation of HMMs uses the Baum-Welch (BW) algorithm. However, the BW algorithm is highly sensitive to initial values of the model parameters. In this paper, we propose an Ant Colony Optimisation (ACO)-based BW algorithm (ACO-BW) for estimating the parameters of HMMs. Our approach benefits from the properties of ACO algorithms and the BW algorithm by combination of both into a single procedure. The improved ACO algorithm provides a new model of artificial ants which are characterised by a relatively simple but efficient strategy of prey search. This is performed by parallel local searches on hunting sites with sensitivity to successful sites. The ACO-BW algorithm also maintains the monotonic convergence property of the BW algorithm. Experimental results show that ACO-BW obtains better values for the likelihood function as well as higher recognition accuracy than that of the HMMs trained by other existing methods.
- 出版日期2011