摘 要: | 1 .INTRODUCTIONBP methodis a widely used classical learning algo-rithmof Neural Networks . Whileit has two short-ages ,i .e .slowconvergence rate and existinglocalmini mum of the cost function. Many new algo-rithms such as modification methods of BP meth-od, conjugate gradient methods , RLS methods ,quasi-Newton methods , evolving learning meth-ods ,chaos-learning algorithms have been proposedso far1 ~7]. However ,these novel algorithms neverdirectly utilize the 2-ed order differentia…
|