LolWolf

35 ms·

331.
by LolWolf 8y ago
10,000! is quite a large number, to be fair. Kidding aside, I agree: I'm also quite dubious of this article's claim on the East India company being worth as much as was computed.
332.
by LolWolf 8y ago
> And putting "simple" before Fokker-Plank equation and "in one dimension" after it is a classic trick straight out of the bluffers guide to sounding like a scientist. Relative to many mathematical tools we touch on a
333.
by LolWolf 8y ago
In general, yes. And even if it's not, people (read: academics) almost universally don't really care and will send you their paper anyways. The main thing is sometimes getting more well-known professors to reply which is a whole
334.
by LolWolf 8y ago
Hmm... both Messiah and Sakurai are usually at the graduate level. If you're fairly strong in math I'd recommend Messiah, but I can't recall off the top of my head whether there is a good chunk of physical knowledge expected.
335.
by LolWolf 8y ago
There are problems, but I’m afraid that we took them down since we need to rewrite a few before next year. Sorry! I’m sure poking around the site might yield some results, if you’re careful enough ;) (at least for 104, someone else is teach
336.
by LolWolf 8y ago
Take a peek at our class [0]! We present a bunch of topics not covered in undergraduate courses (at least to our knowledge), such as proximal gradient descent, random features, etc., without referencing any probability. The course is aimed
337.
by LolWolf 8y ago
Also, if you ever use SQL with the all-caps convention for commands; my right pinky hurts thinking about using shift for all of this.
338.
by LolWolf 8y ago
I think the dynamic is true of many (stable) non-linear systems which operate in a large range (i.e. frequency/bandwidth) of inputs. If there is large noise coming in, the output is bounded, but dominated by the large noise, yet if a s
339.
by LolWolf 8y ago
After the mess that was writing a numerical library that interfaced with scipy but used numpy arrays, I’m actually slowly porting most of my daily workflow into Julia. It’s just gotten faster and the code is much easier to read.
340.
by LolWolf 8y ago
Honestly, just wanted to give a huge thanks for such an awesome language! I’ve truly been converted as a huge Python person into Julia and it’s been slowly taking over my research workflow since it’s just so fast and actually fun!
341.
by LolWolf 8y ago
What about Julia? It’s what we’ve been teaching our Stanford ML class with and it’s fast ! Also typed and JIT with multiple dispatch. Arrays here don’t have the Numpy/Scipy weirdness of matrices vs. ndarrays and linalg is truly first-
342.
by LolWolf 8y ago
The "money" quote is right here, > Steepest descent or gradient descent depends on a choice of ruler (or Riemannian metric) for the parameter space of interest. The Euclidean metric is rarely a natural choice, especially for sp
343.
by LolWolf 8y ago
I mean, if you're already running decent code or already have extremely good tooling where small examples can be easily sent to other machines. Additionally, debugging code running on other machines is also a huge pain in the ass vs. b
344.
by LolWolf 8y ago
For sure! I've managed to get a good chunk of people I work/coauthor with to use it (and like it!), and I've observed an increasing number of my students using it as well (in classes on ML and numerical computing).
345.
by LolWolf 8y ago
In this case, any linear model (with reasonable, minimizable loss, which could be, say convex) will learn the correct thing. Here's a quick gist[0] doing it using least-squares, and learning it exactly (also, for B in the third row you
346.
by LolWolf 8y ago
This is usually quite useful, actually! It happens a lot in numerical computing where you try a bunch of possible solutions and if none are found then you output some sort of "infeasibility" result, otherwise you output the result
347.
by LolWolf 8y ago
Both are true. If there is such a path, then it can be found and is returned, otherwise a verifiable certificate that there is no such path is returned.
348.
by LolWolf 8y ago
Agreed, the example presented in the article isn’t even an (actually useful) application of the original problem. SoS is useful for generating certificates of the stability of controls, rather than generating actual trajectories. That latte
349.
by LolWolf 8y ago
The usual good starting point is Boyd’s Convex Optimization , for a relatively rigorous but also quite practical book. The second usual, quite a bit more rigorous case is Rockafellar’s Convex Analysis , but this one deals only with genera
350.
by LolWolf 8y ago
The article takes a while to get to the heart of the matter, so, even though I do work in this area, it took a solid three paragraphs to realize that this was talking about sum of squares (SoS) optimization. The whole field is interesting b
351.
PCA as a convex optimization problem(guille.site)
3 points by LolWolf 8y ago|0 comments
352.
Persistence hunting(en.wikipedia.org)
2 points by LolWolf 8y ago|0 comments
353.
by LolWolf 8y ago
Yes, that's fine. That Tarski's axioms for Euclidean geometry are consistent is independent of that claim—the above is a statement about the "usual" non-euclidean geometries (i.e., Riemannian).
354.
by LolWolf 8y ago
A somewhat relevant and fantastic read on how large animals have shaped society (and some other interesting thoughts), David Quammen's Monster of God , still remains one of the most fascinating reads I've managed to get my hands
355.
by LolWolf 8y ago
I think there are some crossed wires here. "Consistency" in this sense does not talk about how two systems are related to each other[0]—it is a question of whether you can prove that a system of axioms, call it S , is not absurd
356.
by LolWolf 8y ago
Depending on what you mean, non-euclidean geometry may not be self-consistent (or it may! We don't know, unless we appeal to stronger axioms). In a large number of cases, the set of axioms which produce the geometries in question are
357.
by LolWolf 8y ago
Another fun way: add the two equations to get 2A=11 and subtract to get 2B=9.
358.
by LolWolf 8y ago
I mean, is there any other solution? There are certainly no integral solutions to the problem, but there is a unique (rational) solution. I feel like I’m missing something entirely, though...
359.
by LolWolf 8y ago
There is not, actually! You can show this strategy does as well as the case where prisoners have full information of each other's moves.[0] So any other strategy where each prisoner is blind to others' actions can always be execut
360.
by LolWolf 8y ago
You have to be careful about what you mean by "linear"—linear programs are P-complete, and all binary operations are linear over F_2 (including NAND, etc), since all functions are polynomials over finite fields and the largest pol

More ›