Comment by bonoboTP

Yes. See ICLR for example, which is a peer-reviewed conference, on the OpenReview platform, with no publisher involved, the papers are just hosted on the OpenReview website. It's a similar situation with ICML. NeurIPS also doesn't involve any traditional publisher. Because why would they? There is an online platform to do the reviews, discussions, uploading the camera-ready etc. There is no real role for IEEE or ACM or Elsevier or Springer in this any more. Machine learning as a field has shown that it's perfectly doable, over many years now.