It means the ML algorithm can propose designs that do well on the objective function (e.g. improved runtime), but can't actually be constructed. They give the example of designs that have more memory than can actually fit on the chip.
Yes I understand that, but if that is what is meant I find the wording to be somewhat strange. They mention not being able to find a "feasible mapping from software to hardware", and later on "some of the constraints may not be properly formulated into the optimization, and so the compiler may not find a feasible software mapping for the target hardware".
So the problem is that there is no software mapping, which I understand to be the mapping of compiler instructions to the underlying hardware. It looks like I'm missing something. Is this the same as saying that the hardware design is not feasible?
I imagine they have a basic design in verilog with various tunable parameters (memory size, clock speed, how many instructions to issue at once).
They also have a way to run that hardware in a simulator and see how quickly it could train some network.
The ML optimization problem is to come up with a bunch of constants which performs well, but also compiles into a manufacturable chip. Clearly setting the clock speed to 9999Ghz isn't that...
But what I don't understand is: they claim their approch side-steps the "unfeasible" configs, which is and would be a major achievement, however I don't see how the unfeasibility is captured in their evaluation function, which measures mostly runtime and area, and none of them give negative clear negative rewards to unfeasibility since for example, as you noticed, unbuildable configs would return high runtime... Area might correlate negatively, but at that point I don't see how some methods work (eg evolutionary algorithm) and others really don't... Did you understand that part?
I think you might want to look at something like Solid[0]. It resembles your idea, but is more general. People host their data in a personal data store (a pod, which can be either self-hosted or by a 3rd party) and Web applications read to/write from this data store. It is more general in the sense that this data can then also be used by other applications to provide their own features (which is a hard problem to tackle, since you don't want to restrict all current and future different types of data to one interface).
E.g. When you create a new blog post this is stored in a pod of whichever data provider you chose. The fact that you wrote this blog post can then be discovered e.g. on your social media, after which people can read it in their favorite blog post reader.
I find the implications of such a platform to be the most interesting thing. It effectively creates two different markets: that of data providers, which compete to provide the best service, and of application providers, which compete to provide the best features.
The network seems exciting but it definitely needs a less confusing explanation. Or at least an explanation that does not require prior knowledge of (what seem like) niche concepts.
> "It surfaces content related to the node you're visiting using a lightweight fanout convention"
Requires knowledge of what exactly a "node" constitutes in this context. Since I don't know what such a node is, the remaining part of the sentence about the fanout convention is just meaningless (it also seems irrelevant?).
Now, it is explained on Agora what a node is, namely:
> "A node is the set of all subnodes with a given title, or otherwise mapping to a single entity or wikilink. Subnodes can come from a variety of sources; currently these are mostly notes as volunteered by users via their independent digital gardens."
This suffers from the same problems: I still don't know what a node is, since I would need to know what a subnode is. I'd expect a recursive relation where a subnode is simply a node that is a child of a node, but it seems there is a distinction between both. In any case, it's still unclear.
It seems as if it's written by someone who's so familiar with these concepts that they forgot what it's like to not have this background knowledge. I'd recommend letting someone without prior knowledge explain to you what they think it is in simple terms, and build your explanation from that.
Thanks a lot for your feedback! It is very valuable. I've tried improving [[node]] with your feedback in mind.
Yes, I agree; all text is a continuous work in progress. What you see in the Agora are my notes about the Agora as I develop it and think about it, essentially. Part of the reason I posted this Show HN is to get feedback from people new to the project yet presumably close to the current target audience, if there is any target audience at all :)
I like the idea, in general, but I'm finding this makes for confusing navigation. In particular, the idea of a "node" being the set of subnodes with the same name is counterintuitive to me. Maybe it's just the terminology, or maybe I just find things easier to navigate when they're presented in a strict hierarchy.
I did immediately get the "Everything2" vibe, which is cool, but the thing is -- I've never done anything other than a random walk through Everything2. I couldn't imagine trying a directed or semi-directed traversal of related topics in Agora.
Great stuff. Well witten notes with clear and extensive examples. I actually used this just a few months ago to pass my calculus exam. Feel like I learnt a lot more than from my overly dense and formal textbook.
That's actually an NP-complete problem, called the Dominating Set problem[0]. One application of this problem is used in OLSR[1], a routing protocol used in Mobile Ad-Hoc Networks (MANET). To see how it is useful for MANETs, you can look at the problem in a different, but equivalent, way. Namely, try to find the smallest subset D of nodes (puzzle pieces), such that every two nodes that are not in D are connected to each other through a node in D. In the case of MANETs, these nodes are wireless stations. The nodes in the Domating Set are chosen to relay broadcasting messages for their neighbours. This way congestion on the medium is reduced but every node can still send and receive broadcasts.
It does remind me of the Real World repository [0]. It is basically a Medium clone implemented in tons of different frontend and backend frameworks. Probably not exactly what you were looking for, since it's more related to the frameworks rather than the programming languages, but it is still interesting nonetheless.
From the guidelines on what to submit to HN:
"If you had to reduce it to a sentence, the answer might be: anything that gratifies one's intellectual curiosity."
Don't you think it's insulting to tell a crowd of people that they are not intellectually stimulated by something? I don't find it intellectually stimulating either, but I wouldn't declare something so personal and subjective as factual ("it's not" vs "I don't think it is" or "it isn't for me"). It's antagonistic for no reason.
This is a blog post about a twitter post about a picture. The whole article can be reduced to "tim cook good, trump bad, why in picture together?". HN cannot become reddit, we have to hold our bar higher than blogspam about politics.
It seems a legitimate debate to me to what extent a company should acquiesce to governments of all sorts.
And it's a question that Apple, in particular, has to debate over and over again. iPhone unlocks for the govt? If so, what governments? Allow apps that facilitate protests in Hong Kong? Tolerate or even facilitate oppressive manufacturing or mining conditions?
These are not easy questions for a company that has so many world wide dependencies. And, for some of these questions (unlocks, manufacturing, mining), Apple has decided to take a stand on ethical principles rather than convenience.
Intellectual curiosity is not constrained to geeky stuff.
Corporations, politics, and corporations' political stances and alliances are very interesting, and could be much more "disrupting" and life changing than some new technique or gadget.
Yes, but nowhere is it implied that it is because of something inherent to coffee. It probably could have been another delicious drink, like hot chocolate. That doesn't diminish the content of this article though.
It didn't have to do with the "gathering to enjoy a beverage" (whether coffee or whatever else it could be in its places) either. There were dozens of forces at play to the dissolution of the Ottoman empire. Coffee was hardly any major factor.
First such cities didn't have any lack of places to socialize (and potentially conspire), for milenia before coffee houses.
Second, the same things (nation building, empire dissolutions, revolutions) happened all around Europe and shortly all around the world from the 18th to 20th century, whether they had a coffeehouse culture or not.
The thing with coffee isn't that it's delicious. It's actually quite hard to make it palatable.
But coffee is (1) addictive, and (2) induces wakefulness and action.
I've seen loose theories that coffee was the real reason for the Industrial Revolution. It's not hard to imagine the extra energy it produces to also go into political efforts.
> The thing with coffee isn't that it's delicious. It's actually quite hard to make it palatable
This is like saying it's hard to make chocolate, tea, or many other ultra-popular foods/drinks palatable.
Step 1 of 1: add sweetener. Not hard at all.
Many people (including me) would argue that coffee is delicious without sweetener, though. Perhaps you don't understand how people like it, but that's far from a universal opinion.
My theory/speculation is that if you consume something that tastes badly but makes you feel good, the human body will over time translate the expectation into "this tastes good".
Maybe that's part of how addiction works. Maybe that explains "acquired tastes". And maybe my uneducated guess is completely wrong :)
i never liked coffee. the addiction and wakefulness it induces takes more than one cup. but i was turned off from coffee after just having one sip. so i suspect that there must be more to it than that and that i simply don't understand how people like it.
probably things like mild peer pressure, fitting in with the grown ups, etc, much like smoking, and maybe even drinking.
I don't totally understand what you're trying to say, but again: some people enjoy black coffee, smoking, and alcohol just on their own.
Yes, social factors increase usage of all these things and people will force themselves to do it, but that's not universal.
I also love the flavor of smoking tobacco products (including cigarettes), although I only smoke a few times a year for social, health, and convenience reasons.
the point is, addiction and side effects only take hold after consuming coffee for a while. for someone who doesn't like coffee there is a low risk of getting addicted. therefore there either must be people who do like coffee as it is or they are pushed by some form of peer pressure.
obviously, tastes vary strongly, so both groups must exist.
in my case i escape the peer pressure by discovering that strong coffee actually made me sick, and i used that as an excuse to reject coffee even in situations where it was considered impolite to reject an offer when i was visiting people.
I used to be you. I hated the flavor of coffee, and I also didn't want to be addicted.
Now I've "matured" and realized coffee is a drug. Or, if you will, a medicine. You don't take medicine because it tastes good, but because the effects are, on balance, good for you.
That's funny, I've heard that tea was a potential contributor. The theory goes that the health benefits of drinking tea allowed our British cities to grow larger without disease spreading. This meant more people congregating in one place, which allowed a greater exchange of ideas. Something along those lines.
Naively, it seems to me that as long as the profit you get out of mining scales linearly with the amount of money you put in, the richest player will eventually control all of the hash rate. Yet this doesn't seem to be the case (so far?). Is my thinking incorrect, if so, why?
Well, one thing is that the amount you mine isn’t linear in your hash power, it is proportional to the proportion of the hash power that you control.
Also, measures of how centralized the mining is, is widely believed to have a substantial impact on the price. Miners may believe that if they were to purchase enough mining power to control a majority, that the price they could sell the tokens for would go down, actually reducing their profit. Whether this actually would reduce their profit, I don’t know. Also, conceivably, if they were to buy more mining power, possibly their main sources of competition might respond in kind, resulting in the same income, but higher costs.
By a similar argument to the “avoid there being a majority”, might also want to avoid the case where “if just one large miner drops out, there would be a majority” if people think there is a non-trivial chance of such a miner dropping out.
There is, as I understand it, a relatively small collection of large miners such that together they would comprise a majority. Perhaps this is around the smallest number of independent entities that people will expect to be large enough that they will not collude to do bad stuff, and therefore no miner will buy enough additional mining power to cause this number to shrink, out of fear of making the price go down?
Why would it? E.g. if mining has a 1% return on investment and there are three actors with 4, 5, and 7 units of hash power, and they reinvest profits into more mining rigs, then a year later they will have 4.04, 5.05, and 7.07 units of hash power---the ratios stays the same.
Profit from mining (for example Bitcoin) does not scale linearly with the amount of money you put in. If only a few miners control the hash power of the network, that will discourage use of the currency and drive the price of it down. It's not in the interest of miners to control > 50%.
I'm having some trouble understanding how this manifests itself. Can someone help me with this by e.g. providing a toy example?