ilzmastr

If you ever wondered about the generalization to multiple classes, there is a reason that the gans look totally different:

https://proceedings.mlr.press/v137/kavalerov20a/kavalerov20a...

It turns out 2 classes is special. Better to add the classes as side information rather than try to make it part of the main objective.

How long is your battery life? I've replaced the battery 2 times in my MBP 2015 (iFixit battery) but battery life is only 2.5 hours and I'm perplexed.

Thanks! Where does the "want" come from to compare your "got"s to? If the customer code handles errors gracefully, can you still surface undesirable behavior?

Is there a correlation between recently (last 10 years?) logged areas and wildfires? I watched Wasted Wood by Harvey Richards lately and was surprised to see how much waste is left on the forrest floor, 15-30% of the tree. Seems like it could be an unnatural dense source of flammable material.

The US is the world's biggest lumber producer and Canada is often #2 as far as I can tell.

I'm a woodworker and this movie from 1964 was fascinating. At one point it said 'one day the 16" board will be a museum piece' and I just laughed because it is too true! The only non-composite boards I have seen that wide are $100s of dollars per foot and have such funky grain, checking, cracking, or splitting that I'm sure would have been ripped down by traditional pre-industrial revolution carpenters.

Even in 2021 you could use natural language, not assembly language: https://www.youtube.com/watch?v=Zm9B-DvwOgw

Funny but true, its usually a front page story in the Chronicle if it is going to rain during the week

I'm looking forward to multi-gpu tests!

Would be good to see if it is worth upgrading x4 and x8 setups.

Single gpu upgrade being worth it is a no-brainer. Launch price of the 30xx cards is lower that then purchase-able price of the two comparison cards!

If only you could buy them though. The only microcenter in PA got 15 of each 30xx on respective launch days.

If anyone knows how many of these cards are being produced please do share.

What do advertising-required websites like business insider look like with pi hole? Do they still allow browsing? With ad block on opera for example I am forbidden from browsing with a "it looks like you're using an ad blocker" page.

Great link. The scientist Kevin Knight that did much of the cracking has written about Voynich as well: https://www.aclweb.org/anthology/W11-1511.pdf

As early as 1994, as recently updated as of 04/2020! Great link thank you!

It's not so easy to find a unicode transcription, but I would think that would be the first step in 2020. There is some info here [1] about such a transcription but it does not feel "definitive". Does anyone know if there is a copy paste-able transcription of the manuscript anywhere?

When linear B was decoded the first thing many researchers did was to transcribe every word they could get their hands on, then sort by beginnings and endings of words [2]. I suppose that is what the poster has done since they say there are no inflected word endings, but perhaps a cipher hides these, or they do not have enough data?

[1]: http://www.turkicresearch.com [2]: https://medium.com/@ilyakavalerov/greek-before-homer-how-lin...

This used to exist: https://www.youtube.com/watch?v=CQCgPfw5AMc But Electric Objects is no longer in the hardware business.

Shameless plug, but if you ever need to crop out the frame/surroundings of a bunch of photos of art: https://artsy.github.io/blog/2014/09/24/using-pattern-recogn...

Yea I mean that book basically says that Baum wasn't interested in using his or any algorithm with trading. He just had hunches and a value system that he really believed in for what world events would do to currency prices. I think that part of the book is totally believable. Medallion didn't make a name for itself until long after Baum left.

I highly doubt it. Most of their hires were more machine learning with speech types from IBM, rather than topologists. I doubt the topologists they did hire were being hired for their expertise in fiber bundles rather than for their general intelligence and intellectual curiosity.

I think other commenters are correct when they say HMMs and linear regression made them much of their money in the 90's. I wrote an article [0] summarizing this.

But it is always tempting to think they must be doing something esoteric and mystical at Medallion. A part of me thinks that when interviewed the employees of Medallion say they do whatever simple XYZ technique from quantopian.com/lectures just so that the reporter with a distant memory of HS math leaves them alone. Another part of me does believe that you can do simple things at scale and still make money.

[0]: https://medium.com/@ilyakavalerov/the-man-who-solved-the-mar...

You should state your goals and timeline. There is a lot of deep stuff that can be ignored by many... Often there is a good 1st book, another for 2nd, for 3rd, and hopefully not many more after that.

It also depends what community you ask, engineers vs mathematicians vs actual statisticians. So the community you want to be a part of is the one you should ask. It is probably too much of a melting pot here to get one of these flavors.

In my past I’ve looked at courses I wanted to take, and then looked at the chain of prerequisites and books used there, hws assigned, and did those. Nothing beats talking to the teachers who you think are good. Alternatively you could have colleagues who you like, and whose history you want to emulate.

Having a specific goal is important. Egregious amounts of time can be spent in many corners of probability theory and statistics. Especially in the case that steps are taken in the wrong order. Trying to do Billingsley without first taking real analysis would be crazy. And maybe studying from Billingsley is crazy for most people’s goals to begin with.

That said, I’ve enjoyed reading the suggestions here.

Are you only interested in ML practicum? What did you take after real analysis?

What are your thoughts/interests on analysis for ML, like the approximation theory that branched from Fourier and Wavelet analysis catching on with Cybenko [1], continuing with among others Mhaskar [2], most recently added to by Bolcskei [3]. And then there are other areas where analysis applies like studying the optimization of such networks...

(And of course, this is just for NNs. There are other areas of research where analysis comes into ML. And of course, real analysis lays the foundations for probability and statistics, and is not abstracted away in many research areas in these fields.)

[1](https://pdfs.semanticscholar.org/05ce/b32839c26c8d2cb38d5529...) [2](https://pdfs.semanticscholar.org/694a/d455c119c0d07036792b80...) [3](https://arxiv.org/abs/1705.01714)

tl;dr deep learning needs data to learn invariances, "capsules" build in invariances to 3D rotation... somehow...

"We have the face oval, two eyes, a nose and a mouth. For a CNN, a mere presence of these objects can be a very strong indicator to consider that there is a face in the image. Orientational and relative spatial relationships between these components are not very important to a CNN."

^^^ What? The opposite of this is the mainstream I thought. The promise of DL is to learn hierarchical models of your data. The network learns edge filters, learns combinations of edge filters that differentiate an eye vs a nose, but doesn't learn combinations of intermediate features that determine a face? ppl usually say with a deep enough network an hierarchical concept can be learned...

"We implanted 16/32 site Si-probes in male, adult zebra finches and recorded simultaneously their song and neural activity in HVC; then we used these data to train a long-short-term memory network (LSTM 5) to translate neural activity directly onto song. The goal of the network is to predict the spectral components of the song at a time bin ti, given the values of neural activity features over previous time bins"

This was also food for whiteboards in the show silicon valley. Compare: https://github.com/leandromoreira/digital_video_introduction...

with: http://imgur.com/a/Sne89

On my planet all of those concepts work whether your RV takes values on a countable set (what a discrete RV means) or a continuum.

If the notes only discuss discrete valued RVs there is no problem. This looks like the case since from greping for "continuous" there are few results: https://github.com/ermongroup/cs228-notes/search?l=Markdown&...

This article in the NYTimes is like Duchamp's urinal in a museum.

How is "We carry supercomputers in our pockets. Its simply not ok for apps to be sluggish." not a non sequitur?

Sure they can, many output probabilities: http://deeplearning.net/tutorial/mlp.html#mlp

What is the connection other than "they both have graphs somewhere in them"?

I sort of see what you mean since NNs transport information across a graph in a straightforward (non-loopy) way, and PGMs can propagate information in crazy (neverending loops) ways when doing belief propagation...

How do I submit ideas for other people?

I made http://dillydally.herokuapp.com, a way to share collections of locations with friends. Made it for myself and friends, but didn't get a mobile site done before I had to move on.

Nice work!

How many data points per graph? Really just 1 median salary a year (20x2 data points and 1k^2 pixels :)? I would have preferred just points with straight lines over the interpolated wavy line that looks fibbed. Why not more points? Why not a heat map/2d histogram instead of medians? And why not have the same axis on all the graphs?