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

Self-Classifying MNIST Digits Using Neural Cellular Automata

distill.pub

Post reply on HN
Print view View original

21–27 of 27 posts

Re: Self-Classifying MNIST Digits Using Neural Cellular Automata

#21

Post by SQueeeeeL »

colah3 wrote: post #20
SQueeeeeL wrote: post #19

Earlier quoted context omitted.

I really appreciate the idea of a place where code and paper can be intertwined, it's a cool idea for a journal! I also really enjoy journals doing weird unusually things! I just have a penchant for people defining novel terms and not fully exploring existing literature in a field. In my opinion, it creates a very disorienting effect where those who have useful tools to contribute (typically neural network researcher…

Nope. There's money committed if we ever run the Distill Prize ($10K prizes recognizing outstanding scientific communication), but that's tied up in legal issues. For regular Distill, all the costs (several thousand dollars a year) come out of volunteer pockets, and all the time is volunteered. Edit: There was a period of 6 months where Firebase (which is connected to Google?) gave us free hosting, reducing our expen…

That's impressive, and also re-contextualizes a lot of my feelings towards the journal. Have there been any plans to go more mainline in funding efforts. I know all open access Nature journals are "pay to publish". It's a slippery slope to being predatory, but without university's paying for subscriptions, the funding for a journal must come from somewhere and Nature seems to maintain quality in spite of their model (in this case donations).

Re: Self-Classifying MNIST Digits Using Neural Cellular Automata

#22

Post by colah3 »

SQueeeeeL wrote: post #21
colah3 wrote: post #20

Earlier quoted context omitted.

Nope. There's money committed if we ever run the Distill Prize ($10K prizes recognizing outstanding scientific communication), but that's tied up in legal issues. For regular Distill, all the costs (several thousand dollars a year) come out of volunteer pockets, and all the time is volunteered. Edit: There was a period of 6 months where Firebase (which is connected to Google?) gave us free hosting, reducing our expen…

That's impressive, and also re-contextualizes a lot of my feelings towards the journal. Have there been any plans to go more mainline in funding efforts. I know all open access Nature journals are "pay to publish". It's a slippery slope to being predatory, but without university's paying for subscriptions, the funding for a journal must come from somewhere and Nature seems to maintain quality in spite of their model…

We've sometimes discussed it in the context of whether we can "scale up" Distill. But I think at this point I expect us to just stay niche and volunteer focused.

Re: Self-Classifying MNIST Digits Using Neural Cellular Automata

#23

Post by SQueeeeeL »

colah3 wrote: post #22
SQueeeeeL wrote: post #21

Earlier quoted context omitted.

That's impressive, and also re-contextualizes a lot of my feelings towards the journal. Have there been any plans to go more mainline in funding efforts. I know all open access Nature journals are "pay to publish". It's a slippery slope to being predatory, but without university's paying for subscriptions, the funding for a journal must come from somewhere and Nature seems to maintain quality in spite of their model…

We've sometimes discussed it in the context of whether we can "scale up" Distill. But I think at this point I expect us to just stay niche and volunteer focused.

Based on one of your twitter the calculated IF of Distill is 23. So if you're journal is reaching so many people, wouldn't moving towards more robust and professional editorial staff be beneficial for the large audience? Or do you think there's a benefit too having volunteers?

Re: Self-Classifying MNIST Digits Using Neural Cellular Automata

#24

Post by colah3 »

SQueeeeeL wrote: post #23
colah3 wrote: post #22

Earlier quoted context omitted.

We've sometimes discussed it in the context of whether we can "scale up" Distill. But I think at this point I expect us to just stay niche and volunteer focused.

Based on one of your twitter the calculated IF of Distill is 23. So if you're journal is reaching so many people, wouldn't moving towards more robust and professional editorial staff be beneficial for the large audience? Or do you think there's a benefit too having volunteers?

> if you're journal is reaching so many people ...

Just to clarify, impact factor measures the average impact of a paper, not the total impact of the journal. In fact, it divides by the number of papers! So it's much easier to have a high impact factor if you're very small, and it's probably not quite the right way to think of whether society should invest resources in something.

(In fact, we technically don't have an IF, just the value that would be an IF if we were bigger. IF isn't officially defined until you reach a certain number of publications a year.)

> wouldn't moving towards more robust and professional editorial staff be beneficial for the large audience? Or do you think there's a benefit too having volunteers?

I think there's two main things.

Firstly, it's just a very difficult transition to make. If we wanted to hire staff, we'd need to find people who were aligned with Distill's vision and wanted to dedicate their carer to it. Hiring is hard! We'd also need to find a reliable funding sources (probably at least 200K/year) that was reliable (we don't want Distill to suddenly disappear!) and dosn't impose negative externalities (like being closed).

Secondly, it's a lot harder to experiment when you're big and have commitments to more people. It's already quite hard for Distill to do weird things now -- a lot of people get annoyed to us, and I feel a visceral pressure to be conservative.

Also, a lot of our ability to do weird things and have beautiful articles comes from expert attention to individual papers. A lot of early Distill papers involved Shan Carter or I spending 50-100 hours mentoring authors and giving them design and writing feedback. There's no model where that make sense at scale.

Re: Self-Classifying MNIST Digits Using Neural Cellular Automata

#26

Post by SQueeeeeL »

colah3 wrote: post #24
SQueeeeeL wrote: post #23

Earlier quoted context omitted.

Based on one of your twitter the calculated IF of Distill is 23. So if you're journal is reaching so many people, wouldn't moving towards more robust and professional editorial staff be beneficial for the large audience? Or do you think there's a benefit too having volunteers?

> if you're journal is reaching so many people ... Just to clarify, impact factor measures the average impact of a paper, not the total impact of the journal. In fact, it divides by the number of papers! So it's much easier to have a high impact factor if you're very small, and it's probably not quite the right way to think of whether society should invest resources in something. (In fact, we technically don't have a…

>Secondly, it's a lot harder to experiment when you're big and have commitments to more people. It's already quite hard for Distill to do weird things now -- a lot of people get annoyed to us, and I feel a visceral pressure to be conservative.

I wonder who those people could be, lol.

It's interesting to focus on a work that doesn't function at scale, it seems antithetical to a lot of common knowledge that I often see about the state of the world. I guess this sort of cuts at the same debate which happens at a lot of universities, which is how many students to allow to take a class. I am a very firm believer in having hundreds of niche opportunities to engage instead of one monolithic method of teaching/engagement (and I provide undergrads at my uni the opportunity to work with me and my group for similar reasons).

But it's difficult when the benefits of something are so strong and can lead to direct financial gains, if I spend 20 hours mentoring one UG, they will have a much better chance of getting a good job and every person who I didn't choose to work with won't have that opportunity. Similarly with a journal, every author that's mentored and published is probably getting a straight plus to their career (academic or industry). I don't have any solutions to this, except maybe the complete dissolution of the current state of society, I just find it a difficult circle to square personally

Re: Self-Classifying MNIST Digits Using Neural Cellular Automata

#27

Post by shepardrtc »

lacker wrote: post #25

When I draw an 8 it seems to frequently classify it as a zero. In fact, most of the numbers I draw don't get correctly classified. Just the 0's and 1's are consistent.

I found the same thing. If you draw an 8 with more of an 'x' in the middle, it's fine, but if you draw two o's on top of one another, it can't figure it out. I suspect they need to increase the diversity of their training set.
Post reply on HN