I have watched Video : RNNs (3rd Video) in week 1 of sequence models. In that video, there are 2 problems mentioned with Simple NN because of which we prefer RNNs for sequence datasets.
But I feel that the problem no. 1 (Input & output layer lengths can be different) exists in RNNs as well when I watched the 5th video :Types of RNNs as we are changing the architecture, the same RNNs can’t be used. So, if we want to use Simple NNs, we can change the length of Output layer basis the problem statement which we are dealing. I see the 1st problem is with both NNs and RNNs and not with only NNs. What do you think?
With RNNs, we can change the input and output lengths (the “sequence lengths”), but there’s no easy way to do it with a simple NN.
The layers/architecture of both RNN and NN are not easily changeable once the model has been trained.
To summarize:
A RNN can take in any sequence length, while a NN cannot take in any sequence length.
Both RNN and NN cannot easily change their layer/architecture
1 Like