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

New sphere-packing record stems from an unexpected source

quantamagazine.org

Post reply on HN
Print view View original

111–120 of 226 posts

Re: New sphere-packing record stems from an unexpected source

#111

Post by johnfn »

dkarl wrote: post #12
clickety_clack wrote: post #7

I have trouble explaining to my parents how my job is a real thing. I can only imagine trying to explain ‘I study shapes, but only ones that don’t jut inwards’.

I've found it's best to explain my job using unintelligible jargon. There are three choices, really: You can give a quick explanation in terms they understand, which makes your job sound easy and makes them wonder how anybody gets paid to do it. You can explain what you do and why it's important in terms they understand, but it'll take so long they'll get bored and wish they hadn't asked. Or you can give a quick expl…

The way I think about it is this. There are roughly two groups of people:

- Some people will not care / be dismissive of what I have to say. I probably don't want to talk to these people much.

- Some people will be interested! I probably will like these people.

If I use technical jargon, I am optimizing to impress people I don't really care about impressing - and I will be pushing away the people that I would actually be interested to spend time with.

If I speak respectfully, i.e. the simple explanation, it will resonate more with the people I like. I will push away the people who don't care, but I didn't really want to talk with them anyways.

Re: New sphere-packing record stems from an unexpected source

#112

Post by tbrownaw »

zmgsabst wrote: post #103
tbrownaw wrote: post #77

Earlier quoted context omitted.

Take for example entanglement. The math that describes it is known precisely. Specific implications of this are known. There's no information transfer, there's no time delay, etc. And yet lay people keep incorrectly thinking it can be used for communication. Because lay-audience descriptions by experts keep using words that imply causality and information transfer. This is not a failure of the experts to understand w…

> This is not a failure of the experts to understand what's going on. > We don't have a model of why it works / if there's a more comprehensible layer of reality below it. Counterpoint: You’ve just admitted they don’t understand what’s going on — they merely have descriptive statistics. No different than a DNN that spits out incomprehensible but accurate answers. So this is an example affirming that QM isn’t understo…

The only descriptive / empirical parts is the particle masses.

But it sounds like your objection is that reality isn't allowed to be described by something as weird as complex values that you multiply to get probabilities, so there necessarily must be another layer down that would be more amenable to lay descriptions?

Re: New sphere-packing record stems from an unexpected source

#113

Post by guicen »

YeGoblynQueenne wrote: post #67
dkarl wrote: post #12

Earlier quoted context omitted.

I've found it's best to explain my job using unintelligible jargon. There are three choices, really: You can give a quick explanation in terms they understand, which makes your job sound easy and makes them wonder how anybody gets paid to do it. You can explain what you do and why it's important in terms they understand, but it'll take so long they'll get bored and wish they hadn't asked. Or you can give a quick expl…

I once told my dad that if the subject of my thesis was something I could easily explain then it wouldn't be interesting enough to do a PhD in. I said it half-jokingly and he laughed about it, but he stopped asking me what I'm studying after that so maybe he did take it more seriously.

There’s something bittersweet about that moment when someone you love stops asking about your research. It’s a quiet kind of respect, but also a reminder of the communication gap academia often creates.

Re: New sphere-packing record stems from an unexpected source

#114

Post by zmgsabst »

tbrownaw wrote: post #112
zmgsabst wrote: post #103

Earlier quoted context omitted.

> This is not a failure of the experts to understand what's going on. > We don't have a model of why it works / if there's a more comprehensible layer of reality below it. Counterpoint: You’ve just admitted they don’t understand what’s going on — they merely have descriptive statistics. No different than a DNN that spits out incomprehensible but accurate answers. So this is an example affirming that QM isn’t understo…

The only descriptive / empirical parts is the particle masses. But it sounds like your objection is that reality isn't allowed to be described by something as weird as complex values that you multiply to get probabilities, so there necessarily must be another layer down that would be more amenable to lay descriptions?

That’s not my point, nor close to what I said.

My point is that their models are fitted tensors/probability distributions, often retuned to fit new data (eg, the epicyclic nature of collider correction terms) — the same as fitting a DNN would be.

Their inability to describe what is happening is precisely the same as in the DNN case.

Re: New sphere-packing record stems from an unexpected source

#115

Post by crazygringo »

deadbabe wrote: post #87

Can’t you just montecarlo the hell out of this thing?

According to the article, that's precisely what he did.

But he did it in a better way, randomly expanding ellipses. And when you're dealing with hundreds of dimensions, it's incredibly easy to be montecarlo-ing in the wrong directions...

Re: New sphere-packing record stems from an unexpected source

#116

Post by Dylan16807 »

watersb wrote: post #93
pests wrote: post #42

Earlier quoted context omitted.

Reminds me of the old videos on the Mill CPU architecture. There is multi hour long video about “the belt”, a primary concept in understanding the Mill architecture and instruction scheduling. It’s portrayed in the slides as an actual belt with a queue of items about to be processed, etc. Only in the end to reveal the belt is truely conceptualized and does not formally exist. The belt is an accurate visual representa…

https://youtu.be/QGw-cy0ylCc Is this an asynchronous architecture CPU?

It's not. I'm curious what gave you that idea, though?

The belt moves once per cycle, if that wasn't clear? He says the word "cycle" (and measures latency in cycles) a lot.

Re: New sphere-packing record stems from an unexpected source

#117

Post by Dylan16807 »

zmgsabst wrote: post #114
tbrownaw wrote: post #112

Earlier quoted context omitted.

The only descriptive / empirical parts is the particle masses. But it sounds like your objection is that reality isn't allowed to be described by something as weird as complex values that you multiply to get probabilities, so there necessarily must be another layer down that would be more amenable to lay descriptions?

That’s not my point, nor close to what I said. My point is that their models are fitted tensors/probability distributions, often retuned to fit new data (eg, the epicyclic nature of collider correction terms) — the same as fitting a DNN would be. Their inability to describe what is happening is precisely the same as in the DNN case.

If you have a very small neural network, you can fully understand and explain how it works.

As you increase the detail of a description, it reaches a point where nothing is missing.

Re: New sphere-packing record stems from an unexpected source

#118

Post by CJefferson »

doubledamio wrote: post #17
dkarl wrote: post #12

Earlier quoted context omitted.

I've found it's best to explain my job using unintelligible jargon. There are three choices, really: You can give a quick explanation in terms they understand, which makes your job sound easy and makes them wonder how anybody gets paid to do it. You can explain what you do and why it's important in terms they understand, but it'll take so long they'll get bored and wish they hadn't asked. Or you can give a quick expl…

If you can't explain something in simple terms, you don't understand it well enough

I personally think of this in terms of giving directions.

It's easy to give directions to somewhere near where you currently are -- "Just head down the road, it's the second left, then 3 doors down".

When giving directions to a far-away place you either have to get less accurate "it's on the other side of the world", or they get really, really long. Unless of course they already know the layout of the land -- "You already know Amy's house, over in Algebra Land? Oh, then it's just down the road, fourth left, six doors down".

People often seem cleverer because they know the layout of some really obscure land, but often it's just because people have never been anywhere near it. I have a joke about my research where I say, "A full explanation isn't that hard to explain, it's just long. About 4 hours probably. Are you interested?" So far, I've had 3 people take me up on that, and they all seemed to have an understanding once I'd finished (or, they really really wanted to escape).

Re: New sphere-packing record stems from an unexpected source

#119

Post by tbrownaw »

zmgsabst wrote: post #114
tbrownaw wrote: post #112

Earlier quoted context omitted.

The only descriptive / empirical parts is the particle masses. But it sounds like your objection is that reality isn't allowed to be described by something as weird as complex values that you multiply to get probabilities, so there necessarily must be another layer down that would be more amenable to lay descriptions?

That’s not my point, nor close to what I said. My point is that their models are fitted tensors/probability distributions, often retuned to fit new data (eg, the epicyclic nature of collider correction terms) — the same as fitting a DNN would be. Their inability to describe what is happening is precisely the same as in the DNN case.

... So it's about not being able to observe short-lived particles directly, and having to work backwards from longer lived interaction or decay products? Or about how those intermediate particles they have to calculate through also have empirically-determined properties?

Re: New sphere-packing record stems from an unexpected source

#120

Post by tonetegeatinst »

lupire wrote: post #82
dsp_person wrote: post #66

> For a given dimension d, Klartag can pack d times the number of spheres that most previous results could manage. That is, in 100-dimensional space, his method packs roughly 100 times as many spheres; in a million-dimensional space, it packs roughly 1 million times as many. Those numbers sound wild. For various comms systems does this mean several orders of magnitude bandwidth improvement or power reduction?

I think not, because moving to a higher dimension is exponentially worse (density ~ n^2/2^n) than this linear improvement. So it's only helpful for naturally high dimensional objects. Digital objects do not have a natural dimension (byte length), so you can choose a small dimension. https://en.m.wikipedia.org/wiki/Sphere_packing

The addition of a dimension can be thought of as adding another variable or "axis" in simpler terms.

Due to the added variable aka axis, you have increased the size of complexity.

2d shapes packed in a 2d boundary vs 3d objects in 3d space. The difference is fitting quarters on a paper vs marbles in a perfect cube.

Now imagine having to find the most optimal method of packing for objects of Nth degree in Nth constraint. For example packing object that have 256 dimensional or variables into a constraint that also is 256 of complexity.

I feel as your dimension aka variable increases....the amount of information to compute grows quite quickly.

We can rule out some things as non optimal or non perfect, but we can also get close to perfection via trial and error. I see this as an example of the traveling salesman issue.

Do you stick with a randomly selected answer, do you go with the current most optimal solution, or do you invest time and effort into finding a new solution but you risk finding a worse solution. At the end of the day is the packing efficient gains worth the computational complexity of N dimension of N constraint given it will take an unknown amount of time to find a more efficient packing solution, and the new solution could be anywhere from 0.1% to 80% more efficient.

Post reply on HN