Improvements to context based self-supervised learning (https://arxiv.org/abs/1711.06379)

Deep Learning
1

My first post in this category, so I hope my ignorance does not show too much…at any rate:

Today, at my company, we had a guest speaker from LLNL, Nathan Mundhenk. I was fortunate enough to catch the last 45mins of the presentation. I wish I could have seen/heard the first 15mins because I am lost on the “why” he is doing what is described in this paper. I have tried to read the paper several times but I am still not getting it (and there are no complex math equations!) Even still, I wanted to share the info with the forums for a few reasons:

  1. It was a recent arxiv paper, and I remember reading about this arxiv site in part2 (I never finished, so long ago), so hopefully the information is not too out of date
  2. It looks like there are some topics in here which may help users in their data augmentation techniques
  3. He talked about the “middle layers” in a way, to me, that made it seem as if you could analyze them and just throw one away if the layer wasn’t really helping you. Like seeing that layer4 out of 9 was not improving results, so away it goes! Not sure that comes across that way in the paper.

Anyways, I hope it is of some interest to you. If not, sorry for wasting your time.

1 Like
2

Copying the link here for easier access: https://arxiv.org/abs/1711.06379

3

I wouldn’t worry too much about the paper :slight_smile: I skimmed it briefly and not sure it is the nicest written paper for people who are not super deep in image segmentation. Probably there are nicer papers to read irregardless of whatever you would like to learn more about :slight_smile:

Bottom line is, I wouldn’t worry too much about understanding or not understanding it :slight_smile:

4

thanks for the confirmation and thanks for fixing my hyperlinking ineptitude.

5

Guys, am intersited in Learning more about machine learning and ddeep learning , the technology pix2pix which am totally taken by, AKA image to image transation… i started my frist steps and completed python big courses. any advice where should i start next step ?