Online Time Series Regression about onn

[puzzlecollector]

@alison-carrera thanks for the reply! Yes I guess I wouldn't know until I try. I'll let you know if I get anywhere with my experiments. Thanks :)

[alison-carrera]

Hi @puzzlecollector, nice question :D

It's theoretically possible, if a backpropagation works on an ANN it will work on LSTM as well, but I stay wondering whether it is possible to converge. We know LSTM is a very unstable network and keeps exploding gradients sometimes. Either the online update can cause a lot of gradient explosion or the network will not converge because there are more parameters to be learned. I think you can try this approach, I'm certainly the algorithm will work without executions errors, my only precaution is about convergence. Will it converge? We need to try.