Data Science Stack Exchange is a question and answer site for Data science professionals, Machine Learning specialists, and those interested in learning more about the field. It only takes a minute to sign up.
Sign up to join this community
In order to train a recurrent neural network, you have to unfold it say 50 times and treat it like a chain of RNN cells.
Now if you are using this RNN to train in realtime, then it seems you may as well just have a chain of 50 cells. Maybe fix their weights to all be the same as in a convolutional neural network.
So are RNN's any use in learning in realtime for temporal data?
I think you misunderstood the fundamental of RNN.
fix their weights to all be the same".
Yes that's exact what RNN is suppose to be. In any type of RNN, you always only have one real RNN cell. The unrolled RNN cells are merely representations of that RNN cell overtime. In a single training step, the weights of those unrolled RNN cells are always the same.
Sp are RNN's any use in learning in real-time for temporal data?
Yes, you can train an RNN in real-time with a catch. Although in theory the sequence is arbitrary long, but in practice we can only backprop a certain step back due to
