birdsbolt

Allocations don't have to be expensive if your GC is smart. Smart as C++ destructors positioning :D

Yep, as you acknowledge, your reasoning is irrational. Replacing regular milk with coconut milk is more environmentally friendly, medium-chain fatty acids are better for the liver and heart.

Still, veganism is much more than just a diet and this meaning is what was meant by the vegan-friendly natural latex.

I do assume that people who have never googled "veganism" might have negative opinion of these condoms, but I guess they also won't be aware of the struggles of third world countries and won't care about the goal of this company.

Not that I don't find the fact that veganism had to be mention entertaining.

The author could change the generation of piece by generating a short theme, then fixing that theme at particular points in the note sequence and then generate the gaps, and then repeat the thing as recursively as he wants.

This would provide some recurrence of melodies and would sound infinitely better. Problem with these kinds of models is the window based structure without the notion of the general theme. Theme should then be added by the human, and everything else can be filled using the model.

Why do they need a deep learning model for this? They are obviously targeting signs, product names, menus and similar. Model will obviously fail in translating large texts.

Was there any advantage of using a deep learning model instead of something more computationally simple?

http://musicbrainz.org/ already has this data available.

https://www.youtube.com/watch?v=kW39Mt5kscQ

Another interesting alternative. Prettier than QR and IMO meshtag.

There's also a compiler for Haskell to JavaScript called GHCJS[0]. Many of Haskell libraries can be compiled easily. I think someone compiled Pandoc directly to JS and made a web interface and bundled the app[2].

One interesting things besides web that people have been doing with it is writing Atom editor plugins[1].

[0] https://github.com/ghcjs/ghcjs

[1] http://edsko.net/2015/02/14/atom-haskell/

[2] http://markup.rocks/

This doesn't seem to solve the puzzle. As much as DSL processing part is done, the whole point of the puzzle was to make the operations sublinear in time complexity.

Any generative model can be used to generate a follow up chord. Of course, you need to learn it on a significant amount of data.

There has been some generative modelling of CNNs but I'm not sure if there's generative ability in any of the existing frameworks.

I'm really interested to see how the author will turn Cover Trees into a generic inference algorithm for bunch of models in HLearn library.

Why are archaeologists concluding that this was used for religious purposes? Why is religion always the central theme in those old findings?

The Age of Decadence.

I did chuckle as it reminded me of the "curing cancer" talk Gavin Belson gives in the Silicon Valley series :D

Fundamentally, I guess the difference is that one is type-safe, and the other isn't.

You can get compile time errors instead of runtime for bunch of things.

Sampling should at most take linear time in the number of pixels (ignoring the edge detection). Even less if edge detection result is represented as a sparse array.

Microsoft case has nothing to do with this - even the market share numbers (that are obviously way smaller in case of Apple) do not influence the absence of lawsuits.

Apple produces the hardware and the software.

Microsoft imposed software restrictions (or lets say defaults) on hardware sold by vendors which turned out undesirable for EU. If Apple offered iOS on other phones and restricted the browser application to Safari, they would obviously get sued.

Since they "own" the hardware, and bundle it with their own software - there's nothing illegal about it.

Not that I condone of that kind of behavior, just clearing up the reasoning that was not present in the article.

I think I saw a talk, or a statement somewhere, made by Rich Hickey - https://news.ycombinator.com/item?id=8342718 . I believe it was in this talk.

Not that I was being negative about it.

It improves the performance by fusing some of the operations doing them in one pass instead of multiple passes, and it does so generically - operations are composed regardless of the source, and the implementation isn't looking at the type of the source at all.

I believe one of the arguments was also that these couldn't be written in statically typed languages. Although, I do not know if this turned out to be true.

This is only slightly complicated if you're writing your own transducer operation.

What is interesting is that these transducing operations can be written quite generically - code reuse is huge.

But, usage of transducers is similar to Java 8 Streams, or Haskell stream fusion, only that the implementation is fully generic, not dependent on the container/source.

Large dictionaries (set of words) can compactly be represented with tries.

For words with similar prefixes and suffixes directed acyclic word graph is a much better option (reuses prefixes and suffixes, not just the prefix as in trie), it's a little bit slower to build but fast to traverse if done right.

Any problem where there's a lot of suffix/prefix reusage benefits from a proper trie implementation (or suffix array/tree as alternative) - ex. lempel-ziv compression.

Interesting, so much corn being used to feed animals, so little for humans. A gigantic energy loss trying to raise an animal first, then meat.

It's good to see that efficiency of producing corn has increased enormously.

Author - at the end - proposes a very extreme version of veganism -- cultivating proteins using bacteria feeding them hydrogen, carbon dioxide and oxygen:

"Researchers for decades have been producing food conceived for astronauts on the way to Mars by cultivating hydrogenomonas on a diet of hydrogen, carbon dioxide, and a little oxygen. They make proteins that taste like hazelnut. ... A single spherical fermenter of 100 yards diameter could produce the primary food for the 30 million inhabitants of Mexico City. The foods would, of course, be formatted before arriving at the consumer. Grimacing gourmets should observe that our most sophisticated foods, such as cheese and wine, are the product of sophisticated elaboration by microorganisms of simple feedstocks such as milk and grape juice. Globally, such a food system would allow humanity to release 90 percent of the land and sea now exploited for food."

It's definitely nice to see that all that meat production is getting more and more efficient. The meat industry is still the main producer of greenhouse gases but it's really getting astoundingly efficient, producing more food by producing less greenhouse gases.

There's a lovely book called The Limits to Growth published in 1972, through the years authors have updated their book and their models (there's more than 20). It turned out that business-as-usual model extrapolated very well from '70s to '00s. So, even them modelling was fairly good.

There's no general algorithm, but you could probably prove it, if you tried really hard for your given example. :D

Not really, models in NLP go beyond human performance in some tasks (not tasks as trivial as part-of-speech tagging).

I have a ten year formal training in music - piano (never went to college), I assumed we aren't really talking about composing Rachmaninoff-like pieces. You seem to be aiming at genius-level compositions, that is, currently, unrealistic, and I was surely not talking about that.

You're also going into philosophy of quality. What is quality? Are you doubting the ability of the model trained on thousands of classical compositions to reproduce a fully structured classical piece that sounds well and has a few leitmotifs? It's very easy to constrain the model with a leitmotif positioned at several places and ask of it to find you the most probable sequence (to fill the blanks). It's very easy to take a composition, decompose it into its constituent parts (chorus, verse, etc.) train this kind of sequence to a sequence model, and then do the same for the higher level stuff.

I mean, I agree with you that rule based systems wouldn't work. But statistical models could, if used in music with as much fervor as they are used in tasks in NLP, absolutely produce regular compositions that don't sound like you're randomly spitting out the notes.

Or are you aiming at profound genius compositions? Or maybe super-pop songs? Then I agree, that would be an AI-complete problem, equivalent to machine translation and 300 page novel production.

Of course one needs the knowledge of the first chorus, and that is possible to do with a sufficiently large degree of Markov chain, or you could add all of the notes before the chorus as features during the transitions.

If you use CRFs you can condition on the whole piece and learn the model like that. Yes, you'll have to use a lot of data but models can be as global as you need them to be.

If you want to use a 'verse chorus verse slightly-different-chorus' way of composing, yes, you can use a first level of a chain to generate the probable musical sequence blocks, and generate each block separately, using at the same time features generated in each part (verse, chorus, slightly-different-chorus etc.) to keep the same feeling.

If you train your model in a way described above, you can then pick a tune in your head, put it in and ask the model to generate the most probable sequence for the whole song. Or, if you're using CRFs with Gibbs sampling, you start from the complete piece and iterate until the probability it fits is large enough. Same could be done, somewhat easily, with Hidden Markov Models (I just realised that Markov models might not be the thing I was referring to in the post above, I was talking about statistical variant of Markov chain).

Convolutional neural networks could do the same thing, probably even better than CRFs and HMMs. Music isn't more complex than language and people have been using these sequence modelling methods to do extraordinary things in natural language processing.

Temple of Atal'Hakkar was one of my favorite places (7-8 years ago - vanilla) in WoW. I went there a lot of times, tanking was extremely fun. Silithus, Un'Goro Crater and Swamp of Sorrows were my favorite places for levelling.

Oh, nostalgia is hitting me hard :D

It depends on how exactly global they are, a Markov chain with a large enough degree could be good enough. There's nothing stopping anyone in using additional global features when they're conditioning in the chain. But, that would require quite a bit of data to construct a good distribution. Nonetheless, Markov chains can be pretty smart.

Or, if you want to get even better performance, you can use conditional random fields with exactly the same global features, one has an advantage of not needing so much data because the distribution being modelled isn't a joint one + CRFs are excellent with custom features (features are observed variables and their distribution is implicitly present in the conditional distribution). Disadvantage would be that you couldn't as easily generate the sequence of notes (chords) because the model isn't generative (unlike Markov chains), but one can use Gibbs sampling (combined with CRFs) to search over the space of probable sequences.

Or, some nicely trained convnets could get you even closer to the brain of the composer :D