cscheid

This makes me so happy to read. I had the privilege of working on the same lab as Neil (and Dave Applegate, another notable person in OEIS). No exaggeration at all to call them geniuses, you hang out with them for 5 minutes and know they're cut from different cloth. Nicest folk, too.

Just another upvote for these. They're great if you don't have room for a dumbbell rack, and (of course) they're lighter than the standard dumbbell set.

Yeah, I mostly agree.

I would put it slightly differently, and I honestly don't know if I'm being kinder or not. I'd say that in this book, Feyerabend is being a troll. He's out to get a reaction out of you more than to argue in great faith. In my view it's actually to the detriment of his point.

I'm still happy I read it, but I think it's one of those finicky, "meso-scale" ideas that's useful, but doesn't apply at very small or very large scales. It's also interesting that it came out a good decade after moral particularism came out, and it feels to me that his principle is "simply" methodological particularism.

> embarrassment Kessler syndrome

Thank you, this is a genuinely great turn of phrase.

The values associated with each vertex on the _dominant eigenvector_ (the eigenvector associated with the dominant eigenvalue) are the long-term stable state probabilities. That's from a single eigenvector, not "the eigenvectors".

That's not true. The page rank is read from the eigenvector, and is the value associated with the given vertex (ie web page). There are as many page rank values as there are web pages, but only one eigenvector from which to read: the dominant eigenvector of the transition matrix, which is the one with the largest eigenvalue. So, only a single eigenvalue for the entire pagerank computation.

Antonio Damasio is a neuroscientist who's done work in the area, and has decent popsci books about this idea. I read the older ones, "Descartes' Error" and "The Feeling of What Happens", and they're fun, good reads. Apparently he's written more on the subject as well.

> As such, I hold the entire theory as suspect and someone trying to justify a particular mode of human existence. This extends deeper into the trilogy as women continually are painted as too soft and emotionally empathetic to make the cold hard decisions that will kill many humans but ensure the survival of the race.

For folks who've only read the first book and might be somewhat confused, this specific set of values becomes crystal clear about halfway through the second book.

And also, don't look to get into furniture making where the (previously affordable) good looking, high-quality, easy-to-work-with material is 97% imported from... Russia. Baltic birch plywood, I barely knew ye.

From my understanding, fingerprint analyses don't have widely-accepted specificity or sensitivity rates either. Those "73% match" figures are not derived from any sound principles.

> do you ever shake the feeling that having a book read to you is somehow qualitatively not as good as reading it from the page?

I can't answer for OP, but I did spend a year commuting a bunch and got through ~60 books or so that way.

I went from having time for 10-15 books a year to having time for ~60 books a year, so yes, you do miss a little. But not ~50 books worth of missing. It's easily worth it.

Interesting that you bring up WWII. There's a theory that says that US's big research funding agencies are deliberately set up the way that they are because the US government freaked the _fuck_ out upon realizing that a bunch of nerds could go from zero to nuclear bombs in ~10 years. So, these big bureaucracies were set up to, in effect, keep tabs on physicists. (I first heard of this from a talk from Kim Stanley Robinson of Red Mars fame, and more recently Ministry for the Future. It's on YouTube somewhere but it's a bit hard to find, it was a seminar he gave at Duke University some 15 years ago.)

Then, some 35 years later came the Bayh-Dole act, which however well-meaning it might have been, really provided the incentives for universities to turn into fed money capturing enterprises. The rest is history.

(former tenured prof here)

This is exactly right. Just like it's better to think of McDonald's as a real estate company with a food business on the side, these days it's better to think of big state schools as mechanisms for ingesting federal research dollars, with an education business on the side. (Never mind that state schools shouldn't be education businesses _at all_, they should be _public services_)

(I've been a happy paying customer of Fastmail for _years_ and I didn't know about masked email. Thank you!)

Lowe is being funny and calling back to his classic "Things I won't work with" series (which is also linked to in a different comment).

One of the things you learn just from that series is that anything with this much fluorine jammed into it is just asking for trouble. Case in point, https://www.science.org/content/blog-post/things-i-won-t-wor...

1) Then Donald Logue got famous, and I'd be the idiot yelling "that's Gunter Janek!" at every episode of Grounded for Life.

2) I can't look at that screenshot and not hear "I leave message here on service but you do not call"

3) RSA's Adleman was the science advisor for the movie, so I'd guess he snuck in the Asiacrypt poster from the beginning

I'm sorry, but Datalog has existed since 1977; RDF, since 1999.

Working now, yes. Though deno.land's current IP still nslookup's to something inside googleusercontent.com so maybe google has fixed their Side. But it was definitely offline earlier today (cf. our github CI failures and downforeveryoneorjustme)

https://deno.land down too. I guess a lot of CI actions will be failing just like ours...

If you enjoyed that, you'll probably also enjoy the "Maze Generation" section of Mike Bostock's Visualizing Algorithms talk: https://bost.ocks.org/mike/algorithms/#maze-generation

No, they require a full pass over the support vectors, which are potentially a much smaller set. (That’s part of why everyone was so excited about SVMs when they were invented) The support vectors are the training values with nonzero hinge loss, or alternatively, training values sufficiently close to the decision boundary.

I believe that gives you the result starting from Fibonacci, but not that any starting pair of numbers eventually lands at the golden ratio.

If you read this and you’re curious, there’s a proof that is fairly easy to follow if you understand eigenvectors. Write the operation that takes the two last elements of the sequence and produces the following two, notice it’s linear, then analyze the eigenvalues of the associated matrix and relate the original operation to the power method.

I’m super curious, if you don’t mind my prodding. I looked into doing this for my own version of a JavaScript midi teacher a while back. Any chance you’re doing it via the neural nets sequence to sequence translation algorithms that are super popular in NLP right now? I think that’s a practical way to do it, except for needing a training corpus. I was going to reverse engineer synthesia’s data to try it out, for what’s worth...

I'm not a stats person! But I have been burned in real projects by doing the wrong thing. I just have learned --- the hard way --- to honestly appreciate it.

You make a good point I hadn't considered. In that case, not even robust least squares will save you, because the samples are not independent of one another. You'd have to start looking into explicit time-dependent methods and yikes, I have no idea of what the literature says about time dependence and outliers. I wouldn't be surprised if it's "here be dragons" territory.

As far as i understand it the RANSAC arguments are very much not Bayesian (they're all about "if you repeat this an infinite number of times under an infinite number of new samples, then...", which is almost caricaturely frequentist).

Still, you just gave me reason to mention one of my favorite "no, that won't work either" paper :) on how Bayes will not save you in the presence of model misspecification. Instead of butchering it any further, I'll just point you to this piece explaining the work, written by the author of the paper himself: http://bactra.org/weblog/601.html

Thank you for pointing out some bad phrasing on my part.

When I said "small dataset", I should have said "small subset of all collected points that are initially fed to the model". The issue isn't that you have to collect the data little by little. The issue is that, once you've given a linear model an input that is outside the distribution which you're hoping to model, nothing about the model can be trusted.

So you collect all the data points, but you don't give them all to the model at once. (I'm describing the RANSAC method here now) You start with a large number of "candidate models" that are all fit with a small number of input points, and then test which of the candidate models predict well the points you have not yet given the model. Then you feed the best of these candidate models only the points which it predicts well, and create a more refined, still outlier-free model. This can be proven to work in the presence of a small number of out-of-distribution points.

I don't know how else to put it, but the technique suggested in this paper is irreparably incorrect in the setting that it recommends. Attempting to fix a linear least squares fit by removing points with high residuals does not work. Please don't do that: they won't necessarily be the outliers in the dataset and your model will converge to the wrong thing.

Statistics is hard. Like, really really hard. Stuff goes wrong all the time. Please leave it to the experts.

If you're going to do this, please look up the methods behind (for example) robust least squares, outlier detection, L_1 regression, etc. The right way to do this is to start with a small dataset that is with very high probability free of outliers, and slowly grow it by never adding points which have large residuals. (If you've done 3D scan registration and image alignment, this is what RANSAC does.)

The principle is, intuitively, that once an out-of-distribution point gets into your linear model, the model is poisoned forever. You can't trust the model to tell you that the bad points are the outliers. The way this paper does is is irreparably broken, sorry :/

When I was in a less-than-ideal job situation a few years ago, I found myself saying "... but I love my job" to a friend. I'll never forget the reply: "... you may love your job, but the job doesn't love you back". It was tough to hear it, but in hindsight I'm very thankful for that jolt.