HNHacker News
TopNewBestAskShowJobs

tylerneylon

2,127 karma · joined July 3, 2011

code: https://github.com/tylerneylon personal: http://tylerneylon.com

Email: tylerneylon@gmail.com

submissionscomments
tylerneylon··on The Measurement That Would Reveal The Universe As A Computer Simulation
My impression is that anything you can simulate via a program could just as well be the fundamental laws of physics -- and vice versa.

The idea behind this line of inquiry - which I hope is continued - seems to be assuming that, if we are simulated, then we are simulated using a similar computation model to our own, and using data structures that we would have come up with ourselves. I don't have better suggestions for which model to use, but it's good to keep in mind that even if one model of a simulation fails to match our physics, then there may just as well be another that does match it.

tylerneylon··on Coursera's Functional Programming - Cheating Discovered
Cheating without consequences hurts everyone taking the class. I'm disappointed to see so many comments implying it's fine to just let that go.

My thought on this comes from a good chunk of direct experience: I've taught a number of courses to undergrads at NYU, as well as having about 5 summers of teaching experience in programming and cryptography at summer camps including CTY.

Imagine all students are given a button that says "give me an A," and they can press it if they choose, or do all the course work and possibly get a worse grade. Even if you chose to do the work because you were self-motived, you'd still be tempted to press the A button to guarantee a good grade.

That's an exaggeration, but the point is the same. The easier it is to get _any_ kind of validation without real work, the harder it is to learn.

It is harder to stay motivated, and the feedback received becomes less meaningful. It is even disappointing to simply know many of your fellow students probably cheated without consequences.

I do believe in seeing answers after doing the work, but it is just common sense to avoid posting them before things are due, and to support clear communications between the instructors and students about the details of posting answers.

tylerneylon··on [dead]
Yep, exercise is a key factor.

In the context of this article, by 'serious' insomnia, I meant cases that are not just psychological (eg stress).

I didn't mean to downplay significant sleep troubles due to stress alone. Even in those cases, non-psychological factors should be considered.

tylerneylon··on [dead]
The article is called "Why you can't sleep," and the author most often uses the second person "you." Sometimes the post talks about why "we" can't sleep.
tylerneylon··on [dead]
As someone with serious insomnia, this post comes across almost as selling snake oil. Caffeine, alcohol, general nutrition, sleep hygiene (good sleep habits), sunlight exposure, temperature, eating habits (time/size of meals), all matter -- most of them more than mental state.

From my 20+ years of personally-motivated research on this, I believe the most influential factors in insomnia are, in order:

(1) A controlled, regular schedule of caffeine; limited alcohol. (2) If it exists, a week+ long pattern of waking up at the same time. (3) Exposure to sunlight.

The author's advice is still useful, although I disagree that insomnia "...can be fixed easily and quickly."

tylerneylon··on Shin Mochizuki has released his long-rumored proof of the ABC conjecture
I don't understand the new paper, but here is the ABC conjecture:

Among triples A + B = C, with no common factors between the three, for any given epsilon > 0, there only finitely many with C > R^(1+epsilon); here, R is the "radical" of A,B,C, which is the product of the union of all primes that divide A, B, or C.

The difference between R and the product ABC is that we take out any higher powers of the primes. My intuitive interpretation is something like "in almost every irreducible A+B=C, we almost have C < product(primes(A,B,C))."

Fermat's last theorem (FLT) is a major consequence. Actually, the only proof I've seen connecting the two shows that FLT can have at most finitely many solutions. Here's a useful survey paper:

http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.24.5...

(Click on the PDF icon to download the paper for free.)

tylerneylon··on Accurate remake of Pac-Man in JavaScript
Is it morally wrong that this was created? I appreciate that this exists, and personally am more likely to purchase or promote official Pac-Man products because of it. I think it's a good thing.

Legality is important, but underneath that is the purpose of law itself. I like the idea of keeping in mind both legality and the intent of the law.

If you are making something cool, that adds value (including entertainment), and does not pretend to be, or detract from those you borrow ideas from, it seems ok to me. You may disagree with my particular idea of fair use - the main point is to promote thought about general morality, and not just condemn on the basis of legality alone.

tylerneylon··on Accurate remake of Pac-Man in JavaScript
I enjoy seeing recreations of code like this. It's educational and fun, and this Pac-Man does not take anything away from the original.

It's unfortunate that IP law is so vague and intimidating. It can easily have a chilling effect on innovation and creativity -- you're afraid something might not be ok, but you don't really know, so you just don't try. For me, that's a red flag in a legal system. Competition, creativity, and education are all great things that deserve clarity and protection, both as an original creator, and as someone adding value with the basis of past work.

tylerneylon··on Keep a programming journal
Numbering pages ahead of time sounds like a good way to avoid time-of-writing overhead.

I write down a LOT of my thoughts as I work - most of it I never refer back to again after that day. I like to use book darts (maybe two or three per notebook) to mark things I do refer back to more often. http://www.bookdarts.com/

tylerneylon··on Keep a programming journal
Paper and pen(cil) notes are the ultimate in low-overhead for a temporary close-to-mind data store.

I would disagree about the author's advice to use page numbers and table of contents, though; these add overhead, and for me the main advantage is ease-of-use, including close to zero extra work when writing something down. (Dates are useful, I admit.)

I keep a paper coding notebook of anything I might possibly refer back to ever -- things like basic how-to's on using new technologies, or diagrams of how I'm structuring my code.

I also use a simple open source work journal (nicknamed wj) that helps me make sure I'm on track in the scope of days and weeks. My plan is to end up with one or two sheets of paper that summarize what I did throughout a year - this is a great personal motivator to make sure I'm doing work that a future-me will be happy with. http://tylerneylon.com/a/wj/

tylerneylon··on Twitter acquires Clutch.io
Something similar is appgrok.com which offers server-side control of bundled files, including nib files. It doesn't have a built-in experiment framework, but you could mix AppGrok with analytics to get some experiments-without-app-store-submission working, if that's your goal. (I'm a developer at AppGrok.)
tylerneylon··on Bridges, String art, and Bezier Curves
Is there a place where I can read more about this? Like how to write code along these lines, and maybe how to understand why it works?
tylerneylon··on Bridges, String art, and Bezier Curves
There's a generalization of this problem where we consider the set of all lines between (a, 0) and (0, b) with

||(a, b)||_p = 1

as in the Lp norm; and ask, what curve is produced (as their boundary) ?

And the answer has the nice form: The set of points (x,y) which satisfy

||(x, y)||_q = 1

where q = p / (p+1) [which can be written as 1/q - 1/p = 1.]

This is written up as the solution to a puzzle here:

http://fridaypuzzl.es/?p=187

and if you love puzzles, try to forget what you just read, and here is the puzzle:

http://fridaypuzzl.es/?p=180

tylerneylon··on Amazon Digital Markup: 129,000%
Some commenters say the post is wrong to complain when the author is making money through amazon's distribution, or when amazon takes a loss on selling kindles.

I disagree with both sentiments.

Main point: It does not cost amazon an average of $2.58 to deliver the author's pdf. Amazon is being dishonest by calling this a delivery cost. I consider it unethical to name a fee something that it is not.

"Be grateful for what you do get" is not a valid argument against dishonesty. If Amazon makes a business decision to take a loss on kindles, this also does not justify dishonesty.

tylerneylon··on Ask HN: who started something in 2012 that didn't or won't become profitable?
I built http://thecostofknowledge.com/

This is a place for research authors to support open access ~= the idea that research should be free to read when the authors want it to be. Specifically, the site is about boycotting one particular academic publisher, Elsevier, by pledging never again to do free labor for them (submitting, reviewing, editing papers). So far over 12,000 people have signed up. I see this as part of a larger movement of academics away from research-behind-paywalls.

tylerneylon··on Curse of dimensionality
When it comes to finding approximate neighbors, there is a nice simple trick to help address the curse of dimensionality.

An easy algorithm to find nearby neighbors is to split space into the generalization of squares / cubes, aka hypercubes. Then look at all the surrounding cubes and compile a list of all the neighbors in them. You can do this with a precomputed hash table mapping cubes to lists of points in that cube.

This suffers the curse because a cube in n dimensions has 3^n - 1 neighbors (it is in the middle of a 3x3x..x3 hypercube). Make it a little better with hash table from every cube corner to a list of points in all surrounding cubes, a benefit of 3^n down to 2^n lookups -- still exponential.

Simple idea: Split up space using triangles instead of squares (aka simplices instead of hypercubes). We just went from 2^n lookups down to n+1 (one for each corner) -- pretty good.

The questions are then: (1) how do you tile n-space with n-dim'l triangles/pyramids (simplices), and (2) how do we get highly regular simplices to do that? [Regularity is good because you usually want a sphere-like radius-based lookup.]

It turns out there's a vertex-transitive simplex that tiles any dimension, and as a bonus is easy to work with computationally. ("Vertex-transitive" is a type of regularity where basically every vertex "looks the same" as every other.) So if you care about point neighbors within a certain approximate radius, these tricks are extremely helpful.

Reference: http://www.siam.org/proceedings/soda/2010/SODA10_094_neylont...

tylerneylon··on How to Read Mathematics
I took a course that used this book (Boyer and Merzbach, A History of Mathematics), which I enjoyed:

http://www.amazon.com/A-History-Mathematics-Second-Edition/d...

tylerneylon··on How to Read Mathematics
I agree that typical math writing must be absorbed very slowly and carefully to be understood well.

Unfortunately, in the world of mathematicians, this isn't what really happens most of the time. In grad school classes (for math), students don't have time to absorb the material this carefully - they must 'learn' too much too quickly (or at least this was the case at my grad school), so that reading math becomes learning just enough to pass exams. Writing a dissertation involves a lot of talking to people and skimming papers to hopefully grab relevant bits. Reviewing papers, for professors, involves handing papers to their grad students and asking them to read it. Even writing a paper for journal submission feels primarily about satisfying reviewers and getting in to the best possible journal - being readable becomes a lower priority than space constraints and the quibbles of a particular reviewer.

tylerneylon··on Iterated prisoner's dilemma contains evolutionary opponent dominating strategies
I arrived at some related results in game theory in 2005. If I understand this new paper correctly, section 1.4 in this pdf is similar but less formal:

http://www.math.nyu.edu/~neylon/files/game_thy1.pdf

tylerneylon··on Require free access to scientific articles based on taxpayer-funded research.
To be clear, the 'require' part of this petition is not targeted at researchers, but at publishers.

Publishers want to own their papers' copyrights, and have exclusive publishing rights. The fairness of this is questionable when they aren't paying for the core work done to produce the papers. They apply a lot of pressure to keep research behind paywalls, often against the will of the funders and the researchers. The idea of this petition is to give researchers and government funding agencies some legal strength if they want their work to be freely available.

tylerneylon··on Locality-sensitive hashes are designed to cause collisions and are useful
I created a useful LSH algorithm for real-valued vector data. Its strong suit is that it is deterministic. In other words, it is deterministically guaranteed to find all close-enough neighbors, and omit all far-enough neighbors. Most other LSH algorithms do not provide both of these guarantees deterministically -- instead they strive toward those goals probabilistically. It's also fast, conceptually simple, and built on top of a very cool non-obvious tessellation of arbitrarily high dimensions.

Here are some slides on it:

https://files.pbworks.com/download/dJjN51z5uR/hackerdojo/271...

This work was also published in SODA 10:

https://www.siam.org/proceedings/soda/2010/SODA10_094_neylon...

tylerneylon··on PhoneGap vs. Native: Some Thoughts on Going Native
For A/B testing:

I'm working on a fast testing framework for native iOS apps. I posted some early design specs at appgrok.com. There's also clutch.io, which supports iOS app iterations (things similar to A/B testing) using a hybrid approach, if I understand it correctly.

tylerneylon··on 'Infinity Computer' Calculates Area Of Sierpinski Carpet Exactly
I took a quick look at the arXiv paper behind this article, and it looks suspicious. I didn't take the time to be 100% sure it's vacuous, but it has many red flags, such as being published in "Chaos, Solitons & Fractals," a poorly-regarded journal. Another red flag is the existence of theinfinitycomputer.com.

Now for some real math related to this:

The Cantor Dust and the Sierpinski Carpet really do have area (measure) 0 in the traditional sense. The normal area of the Sierpinski Carpet, during construction, starts at 1 (a full unit square), and is multiplied by 8/9 per iteration. The iterations never stop, so the area goes to zero.

Its 2D measure is 0. There are infinitely many length-1 lines within the final product (any line with equation y=p for any constant p in [0, 1] with no 2's in it's base-3 representation), so it has infinite 1D measure.

So we can look for a dimension between 1 and 2 that measures it, which is where the Hausdorff dimension comes in.

If you double a unit (length-1) line, its measure doubles. If you double a unit square, its measure is multiplied by 4. If you double a unit cube, its measure is multiplied by 8. Here's a pattern we can generalize to non-integer dimensions!

We work from the idea that any measurable set S, when multiplied (scaled) by scalar r, will have its measure (eg length, surface area, volume) multiplied by r^d, where d is the dimension of S. Using m(S) for the measure of S, we can write:

m(S * r) = m(S) * r^d

which means

log [m(S * r) / m(S)] = d * log r

and then

d = log [m(S * r) / m(S)] / log r

This isn't the formal definition of the Hausdorff dimension, but it is the idea behind it, and provides the correct dimension for iterative fractals like the Sierpinski Carpet.

What is d for the Sierpinski Carpet? When we scale it up by 3, we get 8 copies of the original. So m(S * r) / m(S) = 8, and r = 3. So

d = log 8 / log 3 = 1.892789...

for this shape. If you wanted, you could use a factorial-based expression for the size of an n-dimensional sphere, along with the fractional version of !, the gamma function, and use those to get a reasonable number for the exact measure of such a shape.

One more thing. There are numbers of infinite size that you can perform meaningful arithmetic on. My personal favorite are the surreals, championed by John Conway about 40 years ago. This number class basically starts with the ordinals (so including many sizes of infinity) and builds real-number-like operations on those. If omega is the first infinite number, then you can talk about sqrt(omega) in the surreals, and it's a meaningful quantity larger than any finite number and smaller than omega, for example. There are also hyperreal numbers which offer similar freedom of expression.

tylerneylon··on Open, moral and pragmatic
Some researchers are interested in open source, collaborative tools. A great example is the Polymath Project or Polymath1, which resulted in two interesting papers. This project allowed anyone to join the conversation, and de-emphasized the need for personal credit:

http://en.wikipedia.org/wiki/Polymath_Project

This approach is great in that it solves hard problems, which is the real purpose. Unfortunately, it is not much of a career boost, which is a key practical purpose.

Anyone can already publish open source papers on github. I tried this out a while ago, and realized the experiment -- to see what contributions were suggested -- wouldn't be useful without any readers (the topic is a bit esoteric):

https://github.com/tylerneylon/math

tylerneylon··on Tim Gowers replies to Elsevier's open letter
There's a bill that could do this - FRPAA = Federal Research Public Access Act:

http://www.taxpayeraccess.org/issues/frpaa/index.shtml

tylerneylon··on Tim Gowers replies to Elsevier's open letter
New quality journals would be great. Better: new journals with known-good editors, and a way to spread the word that they exist.

There are also emerging platforms that are less traditional, and that I personally like, such as simply posting on arXiv, or things like academia.edu. But it's much easier to move the math world to new journals than to an entirely new system.

tylerneylon··on Math Puzzle: Lion in Circular Cage Puzzle
This is a problem you can be sure you've solved, only to find a twist in the road.

It's problem 1 in The Art of Mathematics by Béla Bollobás, a challenging puzzle book for serious math junkies. The solution in that book has a good explanation of (a) what you probably didn't think of, and sounds right; and (b) why it's wrong.

And just when you think Bollobás's book had the final word, you found out there are versions of the puzzle when both players can win. What? Don't take my word for it:

http://arxiv.org/pdf/0909.2524.pdf

tylerneylon··on The Future of Peer Review
You have some good points.

Perhaps reviewers could publicly vouch for an article to give it credence. Reviewers, who are also researchers, can have scores based on their own papers. If a reviewer with a high score publicly vouches for an article, that article would receive points in favor of being correct.

The most interesting time in writing a paper is when it's under review before being accepted. The process is extremely similar to debugging an app shortly before a release. Part of my motivation for the proposed design is to upgrade this to an ongoing process -- analogous to the move from rare-release desktop software to dynamic web apps. If the author cares about their paper and receives useful comments, then the paper can iterate on that feedback.

I agree that many researchers in practice hear about papers they want to read through word of mouth. For this use case, I like to compare an online platform to stackoverflow (SO), in that you most often visit stackoverflow to solve a focused problem you have, as opposed to browsing for fun (like hacker news). You get an external problem to a page (a math paper / an SO question), you check it out, and you learn what you wanted to learn, or you contribute your knowledge or ask a question. It's a mini-community around a particular idea (a paper or a question). I see this as the main use case. The analogy with hacker news was to imply that the platform could be built around links to arXiv papers in the way HN is built around external news stories (while SO is more self-contained). Like HN, it doesn't seem necessary for an author to submit their own paper to the site.

Finally, I would say that some researchers do browse for fun. I enjoy reading several mathematicians' and computer scientists' blogs, and occasionally visit mathoverflow. This would be a secondary use of the platform, but it would be fun to have a front page of popular papers, and per-topic subpages.

tylerneylon··on What Silicon Valley gets wrong about math education again and again
Learning math from a human in a classroom works because it forms a local culture that cares about math. Students have a chance to be motivated by what the teacher cares about, and this is a very social phenomenon.

It is harder to express sqrt(3/4) on a computer in a meaningful way because there is no ubiquitous and intuitive interface for doing this -- the article has a good point here. But that doesn't mean it's a problem we can't address.

I like the idea of thinking of students in a classroom as young mathematicians, because it respects the students and inclines us to teach in alignment with real-world use cases -- it addresses the 'when will we ever use this' question. From that perspective, it would be a mistake to exclude technology based on its shortcomings. If it is genuinely useful for doing math, then it is worth learning.

I see a lot of concern about math education -- I think most of it misses the main problem, which is that there's zero motivation for people who are great at math to become teachers. If I have a PhD in math, and I become a math teacher, then (i) my math friends will think I will never publish a paper again, (ii) they'll think I am no longer good enough for serious math research, and (iii) I will take a huge pay cut vs typical math-PhD jobs. I might as well add (iv) I would suddenly be working with many people who don't respect or care about what I'm good at (students). I'm an optimist - I think all of these are addressable in the long run. Questions about the virtues of technology are secondary to the ecosystem and cultural perceptions around teachers.

tylerneylon··on The Future of Peer Review
Summary: Current peer review is not 100%; we need to think in terms of authors' motivations (tenure); try a hacker news model, possibly layered with badges-of-approval from editorial board.

The quality of peer review is only as good as the editorial board and reviewers make it -- often this is a lower standard than we expect. This article lists several extreme examples of poor practices by a (previously-)prominent publisher:

http://blog.mathunion.org/journals/?no_cache=1&tx_t3blog...

The key to shifting academic quality control away from the present archaic model is through author's motivation. The replacement needs to be a system that improves and filters incoming papers, and at the same time earns at least as much prestige for the authors who participate.

One idea is something like a combination hacker news and stackoverflow with academic articles in place of news articles / technical questions as the object of discussion. Papers can be changed in response to comments. High scores for either a paper or an author on such a site could replace the prestige granted by acceptance into a top-tier journal. Points are awarded for reviews, and researchers are respected for their scores (closer to tenure), so there is more positive motivation to review than in today's system (which is: it's rude not to review when asked).

I think that alone could work. Some academics might balk at a lack of editorial boards. Fine, let's allow some editors in the mix. An editorial board is an aggregator. They choose what content fits their filter, and may open a discussion with an author on improvements, possibly bringing in reviewers. In the end, readers will see a prestigious badge of the board's approval on high-quality articles, and there could be a page or rss feed to browse the articles meeting this standard. This can all exist on top of the hacker news / stackoverflow system, which itself could be built on top of an arXiv-like system.

Several of these ideas are motivated by thoughts from the blogs of Tim Gowers and Michael Nielsen.

← PreviousPage 5 of 7Next →