selectron

12 ms·

61.
Is There a HN FAQ?
1 points by selectron 10y ago|4 comments
62.
by selectron 10y ago
The best way to reduce overfitting is with cross-validation. The general way is to set up a hold-out sample (or do n-fold cv if you don't have a lot of data) and then use this cross-validation hold-out sample to do feature, parameter,
63.
by selectron 10y ago
Regularization is the process of modifying the model to limit overfitting, for instance by penalizing more complex models. AIC is a specific application of regularization.
64.
by selectron 10y ago
Markets can stay irrational for longer than you can remain solvent. Just because the bubble hasn't burst, doesn't mean we are not in one. Of course the converse is also true, just because the market has been going up doesn't
65.
by selectron 10y ago
Like you said the SM is an approximation, so even with 26 free parameters there is a lot of physics that is left unexplained. In my experience it seems like far more than 26 free parameters which go into MC simulations, for instance with so
66.
by selectron 10y ago
I am not a theorist, but if anything I think the 26 free parameters is a pretty significant understatement in terms of predictive physics. Experimentalists have measured orders of magnitude more than 26 measurements which are incorporated i
67.
by selectron 10y ago
The type of approach used has limited effect on survivorship bias, what matters is the number of people employing different approaches and the size of the effect. So if machine learning approaches can produce real results, the data will sho
68.
by selectron 10y ago
Human-level intelligence is not generally applicable on domains it is not trained on, so holding AI to this standard is ridiculous. Humans need to be taught just like machines do.
69.
by selectron 10y ago
Human intelligence is basically the ability to solve a large collection of sub-problems. As models learn how to solve more and more sub-problems they become closer and closer to human intelligence. Right now the focus is on solving importan
70.
by selectron 10y ago
While overfitting is always possible, in this situation there is an easy causal explanation for why toxic players could be toxic employees. From my experience playing Dota I know that if I was stressed out by work or life I would be more to
71.
by selectron 10y ago
From my Dota experience, usually the hostility is between teammates. Because League partners 5 people together, mistakes and bad play from your teammates can have negative impact on your hero and this can be quite frustrating. Over-extensio
72.
by selectron 10y ago
Yes part of it is certainly hidden information, but for me luck is about how well the hidden information winds up helping you. The "ether around us that makes lucky people lucky" is a bit of a tautological definition of luck, real
73.
by selectron 10y ago
The role of luck explains why your startup succeeds while that of another person who works just as hard and is just as smart fails. Obviously hard work is important, but most startups have a lot of hard-working and smart people behind them,
74.
by selectron 10y ago
Quite interesting. Why couldn't you build in a reward system to using the product? Similar to how games like WOW do?
75.
by selectron 10y ago
Statistical uncertainty grows like sqrt(N), and percent statistical uncertainty shrinks like sqrt(N) / N. So 2000 samples has a percent statistical uncertainty of about 2% - this is likely on par with or smaller than the systematic unc
76.
by selectron 10y ago
As another plug for Kaggle - it lets you know what is state of the art. For instance, from my Kaggle experience I know that gradient boosted decision tress (specifically xgboost) are virtually always superior to random forests. They are als
77.
by selectron 10y ago
Anecdotally as a TA this absolutely matches what I've seen. I think it is cultural combined with courses being much more difficult if you don't speak English well.
78.
by selectron 10y ago
Yes, I am in favor of companies that are based in the US and take advantage of US infrastructure and tax-payer funding (indirectly and directly) should be hiring US citizens as workers. Anyway you didn't address the point - it is compl
79.
by selectron 10y ago
There are always people willing to do the job at the right price and work conditions. Companies would rather pay lower salaries with worse working conditions than pay US workers the wages needed to match the demand. It is basic supply and d
80.
by selectron 10y ago
The don't replace anyone statement doesn't make any sense. Instead of training foreign workers we could be training American workers for these jobs.
81.
by selectron 10y ago
People can't learn things which are truly new from only a few cases of trial and error either. For example, it takes a long time to learn a foreign language or become an expert at chess. Humans mainly learn things better because our tr
82.
by selectron 10y ago
The real question though is whether this is because the model is biased, or whether it is because the two populations have different levels of risk. If group A had a higher risk than group B, than I would expect the model to have a higher r
83.
by selectron 10y ago
It is possible that no significant terror attacks were even attempted against the US since 9/11. And if an attack was stopped they wouldn't necessarily want to publish how they stopped it. So it is impossible to really know if the
84.
by selectron 10y ago
Science is about reaching an expert consensus of the most probably explanation of the available evidence. It is not a democratization of thought - it is evidence that matters not majority opinion or vote of scientists. More generally, it ta
85.
by selectron 11y ago
You are underpaid, but I would not say underemployed. The benefit society gains from a PhD is high, mainly because of the training students get from doing a PhD. But there is an over-supply of students who want to do PhDs as opposed to the
86.
by selectron 11y ago
You are not unemployed when doing a PhD, nor are you goofing off. Not sure if your comment was supposed to be sarcastic somehow, but you are very misinformed.
87.
by selectron 11y ago
Can you give some anecdotes of where "understanding how any important/relevant technology works gives you an unbeatable edge over most of your peers"? Besides being able to pass a job interview. Naively I would assume that be