Backpropagation, Dropout and Simulated annealing

For getting proper weight values, various methods are used. And there is no perfect way.

1. Dropout

It remove a weight randomly and check the result.

2. Backpropagation

In the multi layer model, it adjusts weight values from rear layer. It may be useful because adjusting weight values in the rear layer is more directly related to the result.

3. Simulated annealing.

It modify any weight values randomly. Statistically, this way may be more usuful than formal algorithm.