Ptorch learning record (XIII): recurrent neural network

The basic structure of recurrent neural network

The basic structure of recurrent neural network is particularly simple , Is to save the output of the network in a memory unit , This memory unit enters the neural network together with the next input .
Use a simple two-layer network as a demonstration , On its basis, it is extended to the structure of cyclic neural network ,

It can be seen that the network will combine memory units as input when inputting , The network not only outputs results , The results are also saved to memory cells .
The order of the input sequence changes , Will change the output of the network , Because the memory unit exists , So that after the order of the two sequences changes, the elements in the memory unit also change , So it will affect the final output .