If we have many sequence data and corresponding plots, is there any reason we should use the image rather than the original data for deep learning?

Personally, I don't see any advantage of using CNN on the 2d image data since we could use a 1d CNN on the raw data.

  • $\begingroup$ @MichaelChernick I would guess this question came after he read after this piece. $\endgroup$ – DiveIntoML Sep 1 '17 at 3:53
  • $\begingroup$ Heh. @displayname, if that paper is what you read, I believe it is meant to be taken in jest. $\endgroup$ – Matthew Drury Sep 1 '17 at 4:23
  • $\begingroup$ @MichaelChernick why? $\endgroup$ – stackunderflow Sep 1 '17 at 4:32
  • $\begingroup$ @DiveIntoML I recently join an image recognition competition which uses plots as datasets. I think that's ridiculous, personally. And thanks for the paper. $\endgroup$ – stackunderflow Sep 1 '17 at 4:35
  • $\begingroup$ Its ridiculous in practice yes, but for a competition, it may be just for fun. $\endgroup$ – Matthew Drury Sep 1 '17 at 4:54

No there isn't any reason at all. By plotting the image you are only introducing noise.


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.