诶。。。我也想知道啊。。。

算了给段英文你慢慢理解吧。。。
1 The traditional neural network approach has serious problems to decide when an estimate is close to an appropriate solution and the optimization process can be prematurely terminated.
This is overcome by using estimation with full-rank Hessian matrices of a few selected principal components in the underlying procedure in dmneural
2 It takes a tremendous calculation time for neural network to get an optimized solution for data sets that have a large number of observations.
In dmneural, segments of the data instead of the entire data is trained and the computing time is reduced dramatically
3 Neural network algorithms are prone to finding local rather than global optimal solutions and the optimization results often are very sensitive with respect to the starting point of the optimization.
The dmneural network training can find a good starting point that is less sensitive to the results, because it uses well specified objective functions that contain a few parameters and can do a very simple grid search for the few parameters