The 20,000 for advanced degree holders already holds in the current system, and it already over-subscribed. So this would not reduce the number of H-1Bs issued.
The above commenter probably spoke a little too hastily. "Building Machines" is indeed not a paper about a neural network method, but a survey of problems that we should expect neural networks to do better on, but they currently aren't. That said, the paper isn't down on deep learning, but rather says "we need extra stuff, either additional methods or more inductive biases in our models".
Could you guys elaborate on the relationship between PyText, torchtext, and AllenNLP? I've briefly used the latter two, but with how quickly things are moving it'd be nice to have a quick answer from the devs themselves.
>I grew up in a place practicing one-candidate votes (i.e., you choose between "in favor" that in the end will show 95+% and "against" with no alternatives) so acquired immunity to "citizen votes solve all problems" mentality.
I grew up in a place with effectively single party rule, and it was imbued into our culture that it's pointless to vote because you'll never unseat that party anyway. A big recent change was when, in a semi-recent election, an opposition party won an electoral constituency, worth a mere 5% of parliamentary votes, by the thinest of margins. I couldn't tell you what the opposition platform was today. But regardless of the fact that the opposition still wielded absolutely no legislative power, this led to an era of what many would consider a very electorate friendly legislative push.
This "don't vote for the lesser of two evils" business is nonsense. Use your right to vote relentlessly: to punish arrogant politicians, to press on single issues, to fight for the lesser of two evils because it is the LESSER of two evils. Finding and balancing the lesser of two evils is your job as a voter.
Even if your desired choice has no chance of winning, grinding down the margin, year after year, makes the other side nervous and more willing to compromise. Even if your desired choice has no chance of losing, expanding the margin gives them more room to take less "centrist" stances and push for the things you want. If there's a lesser of two evils, keep voting until the more of two evils has no choice but to compromise and become less evil. Lather, rinse, repeat.
I want to commend you for trying to learn more about the immigration process. More often than not, I find that Americans tend to not know hoops and travails that internationals have to jump through to work in the US, or even just to keep working in the US. Too often, I've heard "H1-B is for cheap foreign labor - just apply for an EB-1/2 or something."
ULM-FiT and OpenAI's Transformer* are quite different. Both are pretrained language-models, but ULM-FiT is a standard stack of LSTMs with a particular recipe for fine-tuning, whereas the OpenAI's Transformer uses the much newer Transformer architecture, and no really fancy tricks in the actual fine-tuning. I suspect the difficulty is with the Transformer model itself - this is not the first time I've heard that it is difficult to train.
* = To be clear, this refers to OpenAI's pretrained Transformer model. The Transformer architecture was from work at Google.
I guess I was a little unclear in what I meant. I meant that everyone (prior to the much later 2DS) had to pay for the 3D tech that most people didn't really want. I absolutely turned my 3D off about a week in and never really used it again.
Just a gamer here, but there's something to be said about the success and failure of the 3DS and PS Vita.
The 3DS, based on hardware/system alone, should have failed. It was tremendously underpowered, and it forced a terrible 3D technology on all its users, which never really took off or became anything other than a novelty while raising the cost of manufacturing.
The Vita was a truly next-generation portable gaming device, with a beautiful screen, incredible graphics, properly analog sticks, an extremely modern interface, and great connectivity and human-interface (camera, capacitative touch) features. Even incorporating the cost of a proprietary Sony memory card, for the amount of power you got, I think it was very reasonably priced. Comparing the 3DS and Vita was like night and day in terms of a modern gaming device. Even today, I think it can stand head-to-head with the Switch in terms of portable gaming.
But Nintendo continued to pour resources into developing top-tier games for the 3DS, slashing its price to bolster adoption, and sticking to its still unorthodox 3D screen/touch screen combo. Whereas for the Vita, Sony quickly got spooked that the Vita didn't perform as well as the PSP, pulled first-party support and general marketing support, and major 3rd party developers (particularly outside Japan) fled the device.
The 3DS is now seen as a major success for Nintendo, while the Vita died (or is still dying) a slow and unceremonious death.
Agreed. I think it comes down to the presentation/interpretation of results. The response to "My classifier gets score of X" can be either "wow, that's a good score for a classifier, this method has merit" or "but X is not a good measure of [actual objective]".
So I think it's come down to conflict between
1. Which the author is trying to present
2. What an astute reader might interpret it as
3. What an astute reader might worry an uninformed reader might interpret it as
And my feeling is that, given all the talk about hype in pop-sci, we're actually on point 3 now, even when the author and reader are actually talking about something reasonable. Whereas personally I'm more interested in the research and interpretations from experts, which I find tend to be not so problematic.
This thread is a microcosm of this whole issue of overhyping.
On one hand, we have one commenter saying he can train a model to do a specific thing with a specific quantitative metric, to demonstrate how deep learning can incredibly powerful/useful.
On the other hand, we have another commenter saying "But this won't replace my doctor!" and therefore deep learning is overhyped.
The two sides aren't even talking about the same thing.
Ah, I was under the impression that RFs choose from a subset of features, not just one feature.
In any case, I agree with the thrust of your original comment that the specifications of the RF algorithm can be relaxed, usually for performance reasons, and still retain strong performance. But this goes back to my original comment that the performance considerations of random forests often aren't highlighted to new learners (whereas introducing ERTs to a beginner would probably shock them - how could you take totally random splits and still get any reasonable performance!)
A funny thing about decision trees (or random forests) is how conceptually simple they are, but in terms of implementation they're very non-trivial.
There's always a point in the lecture or explanation where they go
So we just find the optimal split/feature based on entropy
which no one talks a ton about, but naively implemented is something on the order of O(kNlogN). For each split. Multiply that by the number of leaves (2^depth), and multiply that by the number of trees in your forest.
I learned this the hard way when I tried implementing random forests on GPU for a class (would not recommend: efficiently forming decision trees seem to involve a lot of data copying and shifting around). I actually learned a lot from reading sklearn's implementation of decision trees in Cython - it uses quite a number of neat tricks to make things really fast.
To this day, I'm still trying to understand why my freshman calculus class (proof-based), was as effective as it was.
It was 30 students, 1 lecturer, and a whole bunch of black board panels.
I think part of it was that it felt like a conversation, as well as a game. He would lay out the pieces (assumptions, definitions), and then point us in some direction ("now how would we show X?") We'd throw out ideas if we had any, and he'd either rebut us or nudge us in the right direction. I was 100% engaged in that class - no checking of phones or surfing the net - and it was just myself and my notebook, unlike in other classes that were slides-based and I picked up the bad habit of zoning out when something familiar was being covered. You can't do that in a conversation! The material and strategies I learned in that class completely built the foundation for my math major.
I am still not sure if math is one of the few topics you can take this approach for, slides are the devil, or if the lecturer was just secretly brilliant.
>Facebook run fun psychological experiments to see if they can manipulate their users - without calling for volunteers first or anything I recognise as ethics.
I've always wondered: How is that different from A/B testing, or any other marketing experiment?
That's a trivial but surprisingly reasonable reason for using two browsers. OSX has some annoying quirks sometimes.