1
$\begingroup$

I am exploring 1-D CNN with Keras. My data is $\mathit{k}\times\mathit{N}$ where $\mathit{k}$ is the number of time stamps and $\mathit{N}$ is the number of features. I want to apply CNN with 1-D convolution so that the each filter processes. Can some explain what is meant by dilation in the context of conv1d in Keras (https://keras.io/layers/convolutional/)?

$\endgroup$

1 Answer 1

1
$\begingroup$

Let me first introduce a standard discrete convolution: A discrete convolution

And a dilated convolution may look like this: enter image description here A dilated convolution

When l=1, the receptive field is 3*3; when l=2, the receptive field is going to be 5*5. We could get a larger receptive field at no cost. And this method is not gonna make the kernel bigger, which means we don't need to make more efforts to observe these larger receptive fields.

The receptive field of the dilated convolution

You may read the original article here: Multi-Scale Context Aggregation by Dilated Convolutions

And the following article will help you construct 1D convolutional neural networks: Introduction to 1D Convolutional Neural Networks in Keras for Time Sequences

The differences between convolutional neural networks:A Comprehensive Introduction to Different Types of Convolutions in Deep Learning

$\endgroup$
1
  • $\begingroup$ For dilation in 1-D convolution, will it sub-sample the features or the dilation will be applied on the time-axis only? $\endgroup$ Commented May 12, 2019 at 9:25

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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