bagrow

If you decorate each item with a category tag and then sort using those tags, you are categorizing. The power of decorate-sort-undecorate.

There's a interesting recent paper, also cited by these folks, about errors in gene expression data. Spreadsheets (or at least Excel) are a major culprit, interpreting gene SEP15 as September 15, for example. This is dangerous when it happens in only part of the data, like some columns.

Gene name errors are widespread in the scientific literature

Genome Biology 2016 17:177

https://doi.org/10.1186/s13059-016-1044-7 (Open Access)

Calculus is helpful for understanding backprop.

All the charts are clearly labeled 1991 to 2016.

Your descriptions of their styles (sarcastic-vs-optimistic, whiny-vs-syrupy) are reasonable for their lyrics, but this analysis seems to focus only on the instrumentals, the melodies, chords, and so forth.

Their lyrical differences are reflected in the melodies, of course, but I would not describe Lennon's chord choices as "sarcastic", for example.

Very dissatisifed with the evidence in this article (only national parks?), so I did my own digging.

Here is a 2005 USGS map [1] color-coding nearest road distance for every 30m x 30m square [2] in the lower 48. Surprisingly, it seems to hold up the article's claim reasonably well, and without resorting to discussing 'machines' or 'motors'.

[1] https://pubs.usgs.gov/fs/2005/3011/report.pdf

[2] Edit: the map shows 1 km x 1 km averages of the higher-resolution 30 m x 30 m bins.

Surprised there's no discussion of FFT, power spectra, etc. Would like to see someone with an electrical engineering/signal processing background work on this problem.

The complete Photoshop is a pro-level app that makes a lot of sense on an iPad. Illustrator would also be great. Easy enough to write out a new PDF, but so hard to edit an existing PDF.

Editors sometimes rewrite headlines to make them stickier. Often you can see evidence of the original headline in the article URL (due to CMS issues).

If there is good labeled data, is it still 'hard for people'?

For those interested in some related theory, there is a nice paper on ranking processes in complex systems [1,2].

Also of interest is extreme value theory [3], a cool--but hard!--branch of statistics.

[1] Journal link: https://journals.aps.org/prl/abstract/10.1103/PhysRevLett.10...

[2] Semantic Scholar PDF: https://pdfs.semanticscholar.org/0233/05eda86eb88589b6b28ec4...

[3] https://en.wikipedia.org/wiki/Extreme_value_theory

Who does the grinding? How many people in third world countries have access to trained and equipped lensmakers?

I’m sad about all your hate. I thought posters to HN were more insightful that this.

Any chance of 3D-printing lenses? Frames seem easy, but being able to automate lens creation may be able to bypass these big monopolies.

Whatever happened to these water-filled lenses? https://gizmodo.com/5116969/eyeglasses-with-fluid-filled-sac... (2008)

Anyone aware of research on bandits with a growing set of arms? Meaning, every so often when you play an arm part of your reward is revealing a new arm!

This seems to not fit the criteria anymore (not tabular, not Markov). Is it related to structured bandits?

The paper in question: https://hal.archives-ouvertes.fr/hal-00298818/

Cited over 4800 times according to Google Scholar.

Would this be useful for word vectors/embeddings?

I’d really like to see a propose-and-vote crowdsourced game against AlphaGo. Can 1000 players, with appropriate ranking and selection of moves, perform well? There’s precedent for Chess using basic plurality voting:

https://en.wikipedia.org/wiki/Kasparov_versus_the_World

Great storytelling, but it concludes rather abruptly. An argument of analogy and anecdote. Further, Musk is not the founder of Tesla, Martin Eberhard and Marc Tarpenning are.

I’ve often wondered: Should the number of Nobel prizes awarded scale with the population? There are far more scientists working today than 50 or 100 years ago, and cappping the number of winners makes them more prestigious over time. Perhaps they will become TOO prestigious...

Seems like a stretch to call this “transfer learning”. Maybe training on Sonic and testing on Mario.

Would be cool to see some kind of adversarial competition. You train to, say, beat a game level but you test to beat someone else’s submission. (Short on the specifics, I know.)

This is fantastic!!

Bug report: it appears that multiline summation subscripts are not recognized correctly. For example, Eq. 8 of [1]. These are often created using \substack as part of amsmath.

Awesome tool!

[1]: https://arxiv.org/pdf/1802.01194.pdf

If you turn off smoothing, it really jumps out how noisy and unreliable the signal is that early in the data.

If I may interject a small, shameless self-promotion, we recently published a paper looking at exactly this using GitHub data [0]. Lots of caveats, academic limitations abound of course, but it may be of interest to some.

[0] R Soc Open Sci. 2016 Apr; 3(4): 160007 https://dx.doi.org/10.1098%2Frsos.160007

Accuracy is not a sufficient measure of a classifier. Better to report precision and recall, or any number of other combination measures.

https://en.m.wikipedia.org/wiki/Evaluation_of_binary_classif...

Even the TL;DR on page 2 says it:

> Differences in distributions of traits between men and women may in part explain why we don't have 50% representation of women in tech and leadership

Having submitted to that conference, I really wish they had told authors which reviews came from each arm. Very frustrated with that submission.

Good, important findings for the experiment.

Maybe, but isolating particular trends from overall changes in search interests is challenging.

Here's the trend for "music": https://trends.google.com/trends/explore?date=all&q=music

which is also decreasing. Is music dying?

It's probably not as easy a swap as tensorflow, but stan would be a natural fit for bayesian inference, and its NUTS sampler should be very performant.

http://mc-stan.org

I think the author means a non-ML expert interest in using ML. For example a geologist who wants to apply ANNs her field data

Tunable glasses are great for developing countries as well, where eye doctors are scarce, even when the focus is not automatic [1].

These glasses still require perfectly circular lenses, right? That's a bigger hindrance than slimming down the clunky prototype.

[1] https://www.wired.com/2010/07/fluid-filled-adjustable-eye-gl...