Recurrent neural network (RNN)

One 、 The difference between recurrent neural network and traditional neural network ：

         Mainly introduced successively 、 Time series and other factors that affect the relationship between before and after

Two 、 Network structure ：

         among h0···ht For the results of the intermediate processes , and h0 Act on h1,h1 Act on h2······ It circulates from here , Produce a feedback effect

         The general result is ht

3、 ... and 、 improvement :

        LSTM The Internet ： because RNN The final result is related to all the previous results , Therefore, there may be errors , Therefore, some factors are selectively eliminated , That is, some hn

         Network structure comparison ：

         The specific structure is as follows

         Control parameter ： It needs continuous maintenance in the whole process

         Door unit ： Consistent with traditional neural networks

         Decide to discard information components ：

         Identify updated information components ：