Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Distill: a modern machine learning journal

distill.pub

Post reply on HN
Print view View original

71–80 of 109 posts

Re: Distill: a modern machine learning journal

#71

Post by FractalNerve »

j2kun wrote: post #18

I sure hope this catches on, but we should all be aware of the hurdles: - Little incentive for researchers to do this beyond their own good will. - Most ML researchers are bad writers, and it's unlikely that the editing team will do the work needed (which is often a larger reorganization of a paper and ideas) to improve clarity. - Producing great writing and clear, interactive figures, and managing an ongoing github…

Well, now you need a Distill WYSIWYG, to make it usable (for most of the intended audience).

Hey let's be honest, most academics (that I know) still don't even use LaTeX (or refuse to do so). This is really cool, but requires way too many skills (in js/css3/html5/distill-extensions and node.js).

Personally, my team and I had really great experience with sharelatex.com, whom only I had knowledge about LaTeX. I liked that it's also opensource with a permissive license. I would rather host that on sandstorm.io the next time, or just pay for the comfort offered by overleaf.com (I've never seen such a beautiful colloborative LaTeX Editor).

• What about vendor lock-in?

• Can you export to LaTeX, Word or PDF?

• Can you selfhost it for your team or company?

Re: Distill: a modern machine learning journal

#72

Post by mysore »

Wow this comes with great timing!

I am a UI-developer who has been wanting to learn ML forever. I started working on

1. fast.ai 2. think bayes 3. UW data science @ scale w/ coursera 4. udacity car nano degree

I'm going to write some articles about what I learn and hopefully move into the ML field as a data engineer in 6 months. I figure I got into my current job with a visual portfolio of nicely designed css/js demos, maybe the same thing will work for AI.

Re: Distill: a modern machine learning journal

#73

Post by bpicolo »

colah3 wrote: post #26
TuringNYC wrote: post #25

I don't want to undermine visualizations, they are awesome, but one of the big problems I see with ML research is the lack of re-produceability. I know that Google, Facebook and some others already share associated source repos, but it should almost be mandatory when working with public benchmark datasets. Source + Docker Images would be even better. I worked in clinical research in a past life and studies would be h…

I think that having a venue that can publish non-traditional academic artifacts is an important step for reproducibility, even if it isn't our focus. It seems clear to me that the future will involve some kind of linking reproducibility to papers. If we want to find that future, we need a way for people to experiment with what a publication is.

Jupyter notebooks are a big piece of solving ML reproducability, it feels like.

Re: Distill: a modern machine learning journal

#74

Post by eshvk »

FractalNerve wrote: post #71
j2kun wrote: post #18

I sure hope this catches on, but we should all be aware of the hurdles: - Little incentive for researchers to do this beyond their own good will. - Most ML researchers are bad writers, and it's unlikely that the editing team will do the work needed (which is often a larger reorganization of a paper and ideas) to improve clarity. - Producing great writing and clear, interactive figures, and managing an ongoing github…

Well, now you need a Distill WYSIWYG, to make it usable (for most of the intended audience). Hey let's be honest, most academics (that I know) still don't even use LaTeX (or refuse to do so). This is really cool, but requires way too many skills (in js/css3/html5/distill-extensions and node.js). Personally, my team and I had really great experience with sharelatex.com, whom only I had knowledge about LaTeX. I liked t…

> Hey let's be honest, most academics (that I know) still don't even use LaTeX (or refuse to do so).

What field? TeX is pretty much de rigueur in Math/CS/Physics graduate schools in the U.S.

Re: Distill: a modern machine learning journal

#75

Post by gabrielgoh »

taliesinb wrote: post #56

Great stuff! I'm a fan of what's gone up on distill so far. Question for colah and co if they're still around: When does the first issue of the journal come out (edit: looks like individual articles just get published when they get published, n/m). Also, that "before/after" visualization of the gradient descent convergence is intriguing -- where's it from?

Find out in a week!

Re: Distill: a modern machine learning journal

#76

Post by timClicks »

andrew3726 wrote: post #66
cing wrote: post #19

Is there any concern about a web-native journal being less "future-proof"? I've come across quite a few interactive learning demonstrations in Flash/Java that no longer work.

Also in addition to the sibling comment, the published articles will be on github under their organization.

I feel like binding the journal to GitHub means that it's less likely to exist over the long term (where long term means >100 years, which is as long as I would expect an academic article to be accessible for).

Re: Distill: a modern machine learning journal

#77

Post by colah3 »

timClicks wrote: post #76
andrew3726 wrote: post #66

Earlier quoted context omitted.

Also in addition to the sibling comment, the published articles will be on github under their organization.

I feel like binding the journal to GitHub means that it's less likely to exist over the long term (where long term means >100 years, which is as long as I would expect an academic article to be accessible for).

We produce "archive html" files where everything is bundled into a single file. We're looking into ensuring their long-term preservation with projects like LOCKSS.

Example: http://distill.pub/2016/augmented-rnns/index.archive.html

Re: Distill: a modern machine learning journal

#78

Post by colah3 »

fnl wrote: post #39

How does this provide IF ratings? Probably irrelevant for industry, but publishing in academia is all about IF, no matter how bad and corrupt one might think it is. And what about long-term stability/presence. Most top journals and their publishing houses (NPG, Elsevier, Springer) are likely to hang around for another decade (or two...), while I don't feel so sure about that for a product like GitHub. Maybe Distill i…

Our present JIF is undefined because we haven't existed for two years yet.

If you just apply the formulas anyways, you'll get an JIF of (6 citations)/(4 publications) = 1.5. Again, this number is really pessimistic because those publications are only a few months old and haven't had time to accumulate citations.

> And what about long-term stability/presence.

We aren't particularly tied to github besides it being convenient. Even if the journal died, keeping it up indefinitely would be very cheap.

More than that, we're looking into joining projects like LOCKSS to ensure preservation of the academic record.

> but there seems much to be done to get this really "off the ground" beyond being a (much?!) nicer GitXiv.

We've actually done a lot of the logistics needed to legitimize a journal. We've registered as a journal with the library of congress, joined CrossRef, and built infrastructure to integrate our metadata with the library system.

Of course, there's a lot more to do. But the biggest thing is to just publish great content and run Distill as a serious, high-quality venue.

Re: Distill: a modern machine learning journal

#79

Post by MLasstProf »

j2kun wrote: post #18

I sure hope this catches on, but we should all be aware of the hurdles: - Little incentive for researchers to do this beyond their own good will. - Most ML researchers are bad writers, and it's unlikely that the editing team will do the work needed (which is often a larger reorganization of a paper and ideas) to improve clarity. - Producing great writing and clear, interactive figures, and managing an ongoing github…

Thanks for bringing these points up j2kun.

I'm a junior faculty working in ML with no personal knowledge of web development, d3, etc. While the papers currently on Distill are absolutely gorgeous and will be an invaluable tool for learning advanced ML concepts, I simply cannot see myself or my students putting the time to actually create something like that.

Unless a student is especially adept at the specific tools needed to create these and especially enthusiastic at using them, I will actively discourage them from doing it. The time needed is simply not worth it right now.

I would be happy and grateful if tools for creating these articles become easier to learn and use eventually, such that even the lower-budget, time-constrained researchers could afford to create them.

Re: Distill: a modern machine learning journal

#80

Post by colah3 »

kowdermeister wrote: post #70
colah3 wrote: post #22

Earlier quoted context omitted.

You're absolutely right that this is a lot of work, and not many ML researchers have all the skills needed for it. In the short term, Distill's editorial assistance will help authors produce outstanding papers, although they need to be willing to work as well. In the longer-term, I'd like to explore match making between data visualization people who would like to get into machine learning and machine learning researc…

I already know a guy who's doing this. Although he chose to publish very short videos on various research (including many AI/ML), the concept and goal is more or less the same. Two Minute Papers on YouTube: https://www.youtube.com/user/keeroyz/videos

Karoly does lovely work! :)
Post reply on HN