A study on the convergence of MEBML algorithms
Chuan-Long Wang, Jianguo Zhang · 2002
The mind-evolution-based machine-learning (MEBML) algorithm involving a Markov chain is analyzed in continuous state space. By any finite interval approximation, the convergence of similar taxis operation is proved. Meanwhile, the local property of similar taxis operation is shown. To avoid its prematurity, dissimilation operation need to be introduced. With the concept of absorbing field and p-optimal state, the convergence of dissimilation operation is proved. Finally, the functions of similar taxis and dissimilation operations are analyzed with a view to practical application.