Benchmarking TPU, GPU, and CPU Platforms for Deep Learning

https://arxiv.org/abs/1907.10701

by lawrenceyan • 7 years ago

51 7 7 years ago

Forgive my criticism on a minor point, but that paper's figures are the main content, and the author makes them the size of thumbnails.

If you're going to write a paper like this to share such results, do yourself a favor and make the major conclusions graphically large!

Every one of Figures 8-12 could take up 1/2 a page and not be inappropriate. It's arxiv - what's the benefit of economizing?

piadodjanho• 7 years ago

There is a lot of text commented out in the original files [1]. I reckon they were concerned about the length of the paper. Unfortunately, the original images are still quite small.

do yourself a favor and make the major conclusions graphically large!

Or add the data and the script to plot them in the gzip file you submit to arxiv.

[1] https://arxiv.org/format/1907.10701

_57jb3• 7 years ago

It would be awesome if there were a video channel where folks reviewed these articles in a forum style.

Recognize this likely happens naturally in academia and at conferences but for those of us outside it all, it would (hopefully) raise the level of conversation as opposed to puzzling through some of it.

Likely challenging because so many starting points at different places. I can just imagine a twitch stream with someone reading through providing their thoughts.

partingshots• 7 years ago

I mean it's not like academia is some closed off circle or anything. Especially so within the field of computer science, from the very beginning, everything has always been publicly released and open source on arXiv.

Other fields are quickly improving as well, not to disparage anyone here, with many switching to arXiv in recent years too.

_57jb3• 7 years ago

I wasn't trying to indicate there was a barrier to getting the information. I think the ramp to understanding the information or at least the fact people consume information in different ways it's something I was interested in.

Think more modern-day Homebrew Club. That's what I was getting at

hanrelan• 7 years ago

This youtube channel[1] may be what you're looking for. It has presentations of popular ML papers followed by some Q&A. It seems to target a technical but non-academic audience.

[1] https://www.youtube.com/channel/UCfk3pS8cCPxOgoleriIufyg

_57jb3• 7 years ago

My God. Thank you. I don't know why I didn't think looking this up.

Benchmarking TPU, GPU, and CPU Platforms for Deep Learning

Related Stories