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

Fitting an elephant with four non-zero parameters

arxiv.org

Post reply on HN
Print view View original

41–50 of 155 posts

Re: Fitting an elephant with four non-zero parameters

#41

Post by Steuard »

rsfern wrote: post #14
Steuard wrote: post #11

Sadly, the constant term (the average r_0) is never specified in the paper (it seems to be something in the neighborhood of 180?): getting that right is necessary to produce the image, and I can't see any way not to consider it a fifth necessary parameter. So I don't think they've genuinely accomplished their goal. (Seriously, though, this was a lot of fun!)

They say in the text that it’s the average value of the data points they fit to. I think whether to count it as a parameter depends on whether you consider standardization to be part of the model or not

I see your point, that it's really just an overall normalization for the size rather than anything to do with the shape. I can accept that, and I'll grant them the "four non-zero parameters" claim.

Though in that case, I would have liked for them to make it explicit. Maybe normalize it to "1", and scale the other parameters appropriately. (Because as it stands, I don't think you can reproduce their figure from their paper.)

Re: Fitting an elephant with four non-zero parameters

#42

Post by soist »

visarga wrote: post #40
soist wrote: post #18

Earlier quoted context omitted.

AlphaZero did not create any experiences. AlphaZero was software written by people to play board games and that's all it ever did.

AZ trained in self-play mode for millions of games, over multiple generations of a player pool.

I am familiar with the literature on reinforcement learning.

Re: Fitting an elephant with four non-zero parameters

#43

Post by karmakaze »

jampekka wrote: post #21
maitola wrote: post #16

I love the ironic side of the article. Perhaps they should add the reason for it, from Fermi's and Neumann's. When you are building a model of reality in Physics, If something doesn’t fit the experiments, you can’t just add a parameter (or more) variate it and fit the data. The model should have zero parameters, ideally, or the least possible, or, even at a more deeper level, the parameters should emerge naturally fr…

This was mentioned in the first paragraph of the paper. The paper is mostly humoristic. That said, the wisdom of the quip has been widely lost in many fields. In many fields data is "modeled" with huge regression models with dozens of parameters or even neural networks with billions of parameters. > In 1953, Enrico Fermi criticized Dyson’s model by quoting Johnny von Neumann: “With four parameters I can fit an elepha…

That's how I feel about dark matter. Oh this galaxy is slower than this other similar one. The first one must have less dark matter then.

What can't be fit by declaring the amount of dark matter that must be present fits the data? It's unfalsifiable, just because we haven't found it, doesn't mean it doesn't exist. Even worse than string/M-theory which at least has math.

Re: Fitting an elephant with four non-zero parameters

#46

Post by Sesse__ »

aqme28 wrote: post #34

> It only satisfies a weaker condition, i.e., using four non-zero parameters instead of four parameters. Why would that be a harder problem? In the case that you get a zero parameter, you could inflate it by some epsilon and the solution would basically be the same.

They also, effectively, fit information in the indexes of the parameters. I.e., _which_ of the parameters are nonzero carries real information.

In a sense, they have done their fitting using nine parameters, of which five are zero.

Re: Fitting an elephant with four non-zero parameters

#47

Post by pharrington »

soist wrote: post #42
visarga wrote: post #40

Earlier quoted context omitted.

AZ trained in self-play mode for millions of games, over multiple generations of a player pool.

I am familiar with the literature on reinforcement learning.

They're saying the board games AlphaZero played with itself are experiences.

Re: Fitting an elephant with four non-zero parameters

#48

Post by edflsafoiewq »

maitola wrote: post #16

I love the ironic side of the article. Perhaps they should add the reason for it, from Fermi's and Neumann's. When you are building a model of reality in Physics, If something doesn’t fit the experiments, you can’t just add a parameter (or more) variate it and fit the data. The model should have zero parameters, ideally, or the least possible, or, even at a more deeper level, the parameters should emerge naturally fr…

Isn't the form of an equation really just another sort of parameter?

Re: Fitting an elephant with four non-zero parameters

#49

Post by qarl »

edflsafoiewq wrote: post #48
maitola wrote: post #16

I love the ironic side of the article. Perhaps they should add the reason for it, from Fermi's and Neumann's. When you are building a model of reality in Physics, If something doesn’t fit the experiments, you can’t just add a parameter (or more) variate it and fit the data. The model should have zero parameters, ideally, or the least possible, or, even at a more deeper level, the parameters should emerge naturally fr…

Isn't the form of an equation really just another sort of parameter?

Yes, it is.

Which makes the only truly zero parameter system the collection of all systems, in all forms.

Re: Fitting an elephant with four non-zero parameters

#50

Post by edflsafoiewq »

karmakaze wrote: post #43
jampekka wrote: post #21

Earlier quoted context omitted.

This was mentioned in the first paragraph of the paper. The paper is mostly humoristic. That said, the wisdom of the quip has been widely lost in many fields. In many fields data is "modeled" with huge regression models with dozens of parameters or even neural networks with billions of parameters. > In 1953, Enrico Fermi criticized Dyson’s model by quoting Johnny von Neumann: “With four parameters I can fit an elepha…

That's how I feel about dark matter. Oh this galaxy is slower than this other similar one. The first one must have less dark matter then. What can't be fit by declaring the amount of dark matter that must be present fits the data? It's unfalsifiable, just because we haven't found it, doesn't mean it doesn't exist. Even worse than string/M-theory which at least has math.

It's easy to say "Epicycles! Epicycles!", but people are going to continue using their epicycles until a Copernicus comes along.
Post reply on HN