T_D_K

I understand all that, which is why I said the industry has room for improvement. What I'm arguing is that they have value -- they save you from taking on what is essentially a part time job that you're going to be bad at, and that they have potentially years of experience in. Whether the way in which they're compensated is fair or not is a different argument.

Having a real estate agent as a buyer can be extremely useful. Time savings: they'll actively search out properties that fit your needs, handle negotiations, schedule inspections and work with the city or county, etc. They're also experienced with all of the legal aspects, and work with mortgage and insurance companies.

"Tech Bros" have a stereotype of thinking they can do everything themselves. Sometimes they should maybe step back and acknowledge that someone's profession is a bit more than what they can pick up in the course of reading blogs for a few weeks.

Now, it's understandable that there's plenty of bad real estate agents -- finding a good one can be difficult. And I'm not saying that the industry is perfect, there's certainly room for improvement. I just think that it's unfair to say they don't have value.

*ungood

I'm not sure how that's a big deal for Lyft. It seems like a pretty healthy supply and demand curve -- if the number of drivers drops, it becomes more profitable to be a driver. And there's always people looking for supplemental work, who'll go where the money is. I suppose the part I'm missing is the reduction in riders if there's not enough drivers, but that is apparently not a huge issue (according to their rider-retention numbers).

Do you disagree?

Are you in a union? Your description sounds very similar to my father's experience in the USPS. If so, are the problems you describe related to the fact that it's a government job, or the fact that it's a union job?

That's noble, but many aren't inclined trade personal profit for the company's profit. I don't think they should be, either. It's the company's job to ensure that their success is aligned with their employees'. Raise the issue to management, and if no action is taken, then you've done your due diligence.

I can imagine situations where you should care about the company's performance, like a non-profit or certain startup situations, but I think the majority of enterprise developers fall under the above.

From the article: "The app brings up the (scanned person's) vital information, including name, ethnicity, gender and address (..), the address of the hotel where they are staying and information related to their internet usage."

So, yes, every person is tracked.

Ah, thanks. The decibels example makes sense (an alternative would be to take the log first, and then convert it back after averaging?), and I can see how the 0-10 system can also be viewed as categorical rather than discrete.

I've never heard this before, and a skim of the wiki page doesn't mention it as a prerequisite. Mind explaining? The scores are just integers, so the addition is well defined. So you're saying that the context is what's relevant?

Not that the mean is the only (or even the most useful) statistic.

Great explanation, thanks.

And the "distinct eigenvalues" part is obvious in hindsight. For some reason my brain thought that we were adding them, not subtracting.

I got lost in 2.2, I can't work out how applying the transformation leads to the result. Which is frustrating since it's the only non-trivial line in the proof, lol. Also, after applying the transformation, the author states that "a1(λ1 − λ2)(λ1 − λ3)...(λ1 − λm)v1 = 0" => "a1 = 0". But he never says why we know "λa != -λb for all a, b in 1..m" -- that seems non-obvious to me.

Well, all of that is really just Hungarian Notation [0] with typesetting rather than extra characters, so it's not 100% necessary to fully understand -- just look at the definition again. The exception is standard sets, like the Reals or Natural Numbers etc. (which use the double strike through). But that's pretty standardized.

[0]: https://en.wikipedia.org/wiki/Hungarian_notation

>But this wasn't the case in later Rome and Europe, not even in the Inquisition era

I'm skeptical that throughout the history of Christianity, everyone took the "parable" interpretation of the bible (rather than taking most or all of it seriously). Are you claiming that the... inquisitors (?) weren't killing based on a literal lack of belief in god, but rather were killing because people weren't morally up to snuff? I'd love to be educated here, theology and the history thereof are well out of my wheel house.

Personally, it seems reasonable that "Jesus" was a Joseph Smith type of con man who managed to spin yarns and somehow create a following.

Maybe I'm slow, but I still don't get it. Is he saying that members of the state's church will follow him blindly (making them out to be dumb / sheeple)? Or is he saying that even such a ridiculous sight is less ridiculous than the church? Maybe he's satirizing the state's enthusiasm for war? I feel like I'm missing context here. Or maybe this is why people say his humor is weird...

Based on the limited amount of information, I'm assuming that by "training explodes" you mean that your gradient descent never reaches a local minimum. Try lowering your learning rate? You may be "stepping over" the minimum.

This is a great response! But I think the person you replied to understands all this, and is wondering about the exact method (or a layman's explanation) of the same process in terms of elliptic curve crypto rather than the traditional RSA (I'm wondering the same thing).

First question: The network is trained to recognize a fixed set of outputs. That's what makes it a classifier -- it classifies an input into a single output. It does this by giving each output possibility a score, and the highest score is its guess for what the original image is. So if I have a network that I train to recognize cats, dogs, and cars, and I get an output like {cat: .13, dog: .85, car: .02} Then the input was most likely a dog. The network calculates all of those values simultaneously.

You can, of course, tell the network to output whatever you want: all of the guesses, best guess, top five guesses, all guesses over a threshold, etc.

Note, this is a gross oversimplification, but it gets the general concept across.

These are called hyper parameters (filters, filter sizes, stride length, pooling function, activation function, and a whole host of others not mentioned in this article). They are chosen "randomly" in the sense that it isn't an exact science, ie there is no "right" answer. However, intuition and experience are used as a guide to select reasonable values.

The values in the filter matrices and the weights and biases of the fully connected layers are truly random though. They are often initialized with Gaussian random values. Sometimes they are just initialized as all 1's, or 0's. Again, there's no "right" answer (there is probably research out there that recommends one initialization approach over another). These are the values that are trained using gradient descent.

Ok. I stared at it long enough, and I think I understand. Being Lipschitz-continuous means (in a non-rigorous way?) that a the gradient / slope of a function has an upper bound. And Hessian-Lipschitz means the same, but for the second derivative / hessian.

So, f(x) = x^2 is not Lipschitz-continuous (because the slope gets arbitrarily large), but something like f(x) = sin(x) is Lipschitz-continuous because the slope never exceeds some upper bound.

Funny how trying to write down the question gives the brain the kick it needs sometimes :)

I'm getting stuck trying to understand the equations in assumptions 1 and 2. Can anyone point me to a resource that explains the idea behind them? Wikipedia is a bit terse, and I'm not having any luck googling for "gradient-lipschitz and hessian-lipschitz" and variations.

On the notation side, am I correct in thinking that "<del>f(x_n) is the partial derivative w.r.t. x_n? And that the elements of the vector x are the parameters against which a "cost function" (f) is computed? But that doesn't seem right. Maybe x_n is a point in R^N, and therefore <del>f(x_n) is the derivative at that point?

Does anyone have insight as to why they're still doing top 5? It seems to me like the error rates have dropped low enough that they could move on to top 3 or even single guess challenges. Is there data that shows how these same models perform in such tasks? Though I suppose, if I was motivated, all the needed tools are available to find out for myself.

He's proud of what he's accomplished, in a weightloss post-mortem. How does that make him narcissistic?

It's easy to maintain your goal weight if you truly enjoy the lifestyle you've created. Obviously, if you get up one day and say, 'Well I don't actually like going to the gym / tracking calories / etc., so I'll skip today', you've started to gain the weight already. The trick is finding a healthy lifestyle activity that can keep you active and that you can stick with. Climbing, running, hiking, kayaking / canoeing, cycling, tennis, etc etc are all great options if going to the gym to lift weights isn't your thing.

Why does the article say there are 2^N different possible configurations? Shouldn't it be N! ? Maybe I'm missing something, but I don't see how they can rule out so many different orderings of the "configurations", since he started out by saying that the ordering wasn't reliable.

Edited: I figured it out. It's because they essentially do a binary search of the changes to find the issue, I should have read a bit further before commenting. The tables / examples cleared it up pretty quickly.

Wow, thanks for the resources!

> I grew up playing soccer, and even there they, at least in the US, as I understand it from my nephews, heading the ball is illegal until high school.

Maybe in a Rec league, but if it's any level of competitive play then they'd have been heading the ball since grade school.

Looks great so far! A question, though: why are the group sizes limited to 5 people? Is it an issue with the cross referencing being too strict to find good matches with more people? I would like to have a larger group size, 10 would be ideal, though even just 7 would be a big help.

I suppose the author has an interesting point at the core, but the article is way too prejudiced to be taken seriously.