There's already a decent discussion on how to select the right number of hidden layers and hidden nodes in a feed-forward neural network: How to choose the number of hidden layers and nodes in a feedforward neural network?. However, I struggled to find a detailed discussion on how many hidden layer nodes LSTMs, GRUs or vanilla RNNs need to perform well. We do have some discussion online on other platforms, for example: https://ai.stackexchange.com/questions/3156/how-to-select-number-of-hidden-layers-and-number-of-memory-cells-in-lstm, but answers, on average, seem to be at least about a year old, and given the speed of deep learning research, that may mean they're outdated.

The task I'm currently working on is predicting a single binary outcome from a fixed-length multivariate time series, so we may as well fixate on this specific context of using RNNs.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.