Intuitively, are both RNN and 1D conv nets more or less the same? I mean the input shape for both are 3-D tensors, with the shape of RNN being ( batch, timesteps, features) and the shape of 1D conv nets being (batch, steps, channels). They are both used for tasks involving sequences like time series, NLP etc. So my question here is this,

Are the steps and channels in 1D conv nets similar to the time steps and features in RNN? If they are, then why don't we use Conv 1D for time series problems instead of RNN since they are much faster compared to RNNs?

Please note that this is not a direct comparison, I know that they both work differently on an architectural level but I am just trying to get a high-level overview.


Yes the interpretation of the dimensions is pretty similar in both cases.

An important case where RNNs are easier to use is with data of unknown lengths. For example, in sentence translation (e.g. translating Chinese to Icelandic) both the input and output sizes are dynamic. In this case, it is easier and more intuitive to use RNNs than to try and shoehorn in CNNs.

If instead your task is closer to classification of fixed length sequences your intuition seems correct. In my experience, usually 1D CNNs are faster to train and perform better in this case.


Your Answer

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

Not the answer you're looking for? Browse other questions tagged or ask your own question.