« first day (2271 days earlier)      last day (3042 days later) » 

23:00
I think foot pounds are called stomps.
@Axoren Y'all needda switch to metric man :P
@Danu I agree. There are so many measurements I use in my daily life that make me wish everything else was measured in metric.
Metric ! Metric ! Metric !
I measure my water consumption in liters, but need to measure milk by the gallons.
@Danu The thing is that for hardware in particular most nuts bolts etc are made in imperial, so tools that are built in metric are not helpful - they're often just slightly off in size. And while we can pretty easily just change the signs with milage info, it's a lot harder to change the physical reality of things that exist and are in your home.
23:04
@MikeMiller I've heard that the metric tools and the English tools are actually the same as each other in plumbing and that none of those measurements are actually present anywhere on the pipes or tools themselves. A quarter-inch pipe is the one that corresponds to the quarter-inch fittings and nothing more.
Funny how the conversation degenerated while I was working out equations of projective conics.
It sounds like a plausible horror story.
@MikeMiller Hmm
I don't know about plumbing, but with nuts and bolts they do correspond correctly to measurements.
Okay Good night Everyone :D
23:06
I'm glad they do. We ended up getting the metric set and one of the wrenches managed to fit snuggly in a eighth-inch screw.
Ended up being the right conversion when we checked it after the fact.
Bye.
Bye, @Mahmoud.
What doesn't make sense to me is diagonal measurements for television screens.
is that periodic?
Is $\delta$ the Dirac delta?
23:07
yup
Horrifying.
or as its known in my world, the unit impulse
So you have impulses at $3k$ and $k^2$ for all possible $k$.
Is $\sum \delta[n-3k]$ periodic?
thats just 1 right?
23:09
isnt the impulse (1) where you define it?
Yes. But there's no impulse at 2.
There is no integer $k$ such that $3k = 2$
Nor is there one such that $k^2 = 2$
The answer to my question is YES. The answer to the original question is NO.
k yeah
that makes sense
since the second one is (1) everywhere
23:11
Huh?
you can keep iterating k
whereever k^2 = 3k you get (2)
Again, no, it's not 1 everywhere. It's 1 at the points that are squares.
right ya
alright thanks guys lol
Good luck with engineering school.
You should really try playing with examples.
23:13
As Danu said, graph stuff out on paper.
Drawing dots for the sum of unit impulses would take much less time than summing sines
You can check that x[0] = 2, x[1] = 1, x[2] = 0, x[3] = 1, x[4] = 1, x[5] = 0, x[6] = 1, x[7] = 0, x[8] = 0, x[9] = 2...
hey @TedShifrin just want to verify with you something. Suppose we consider the sequence $l_{\infty} = \{ x = \{x_n\} : sup|x_i| < \infty \}$. Suppose $\{c_i\}$ is a cauchy sequence in this means that for every $\epsilon > 0$ there exists $N \in \mathbb{N}$ such that $i,j \geq \N$ means $sup_{k \in \mathbb{N}}|c_i^k - c_j^k| < \epsilon$. So in particular we have for all k and $i,j \geq N$ $|c_i^k - c_j^k | < \epsilon$. This means that for all k $\{(c_i)^k\}_{i = 1}^{i = \infty}$ is cauchy in R.
By completeness of R we have that for such sequence limit exists. I.e $lim_{i \rightarrow \infty} (c_i)^k = c^k$. Let $c = (c^1,c^2,.....)$
Your notation needs work, Karim.
now I am having troubles showing that this $c \in l_{\infty}$
23:16
Better not to use exponents when we actually take powers sometimes.
what would you suggest using ? I am just following differential geometry notation
Differential geometry notation is not good when you have powers of variables. I would let $c_{ij}$ be the $jth$ element of the $i$th sequence.
@Adeek I'm partial to $a_{b,c}$ when I have to deal with such an indexing problem in my work
@Axoren yeah I guess I could use that notation.
Notice I did the same thing without the comma.
23:18
The comma lets you tell the difference between $a_{121}, a_{1,21},$ and $a_{12,1}$
So why don't you write $a_{i,j}$ for matrix entries? :D
haha
I do, personally.
In diff geo you would be strung up. The comma is used to denote I'm (covariant) differentiating with respect to the $j$th variable.
@Axoren It's funny that you say partial, because in physics $f_{a,b}$ sometimes means $\partial_b f_a$, in differential geometry.
23:19
Seeeeee?
oh haha
$\ _b^an_{c,e}^d$
This is our solution.
haha
Even better now.
hmm
23:21
Anyhow, Karim, you seem to be saying that if you think of sequences as functions defined on $\Bbb N$, that you have pointwise convergence to $0$.
@TedShifrin Awful.
Why awful?
Apparently, $\ ^an$ is taken by tetration.
I don't even know how to LaTeX that, and nor do I care to. :)
Yes. I want to first show that such constructed sequence that I have from completeness of $\mathbb{R}$ is first in $l_{\infty}$ and it converges the cauchy sequence converges to it.
hm just a sec I think I have an idea why it is in $l_{\infty}$
23:23
\$\ ^an\$
I think the trouble that I am having is with my notation.
See what I mean, Karim? :D
Literally the first think Ted said to you, lol
yeah haha
Karim: Seriously, it might clarify things in your head if you think of them as functions with domain $\Bbb N$ and write $f_i$ for the elements of the sequence.
23:25
yeah the troubles is with my notation I will write a new solution with new notation.
yeah I will do that @TedShifrin
I know this isn't the programming stack exchange, but the query is somewhat mathematical: do any of you know standard terminology and/or writing syntax for papers about convolution neural networks?
Don't look at me.
Ugh, there is a standard terminology, but I don't deal with convnets enough to know it.
People also regularly deviate from the standard.
Buzzwords to look for are Filters, Convolutions, Correlations (180-degree flip of the convolution operation in the context of image convolutions)
Beyond that, I can't be much help without a specific terminology question
do you guys use convolution in the same way we do?
More often it's not in the discrete setting, but same idea, doomedPing.
23:31
where its the integral of product between a function and infinitely shifted impulse?
There are a lot of contexts in which its used, but I think they're generally the same idea.
continually shifted rather instead of infinitely
You don't have to convolve with an impulse. You can convolve two signals.
yeah i guess
sorry i just meant in terms of the TF. my bad
yeah thats true
how do convolutional neural networks work?
In convolutional neural networks, my understanding is that you're applying a filter to local regions of an image and this filter application is called a convolution.
23:33
oh?
thats interesting
We did a very brief bit about it in my AI class
i didnt know that it was an image processing idea
i thought it was only AI
Look up Computer Vision
One of the big fields in AI
Hell, 95% of Photoshop is the result of AI and Datamining
thats really cool
signal processing meets AI
I thought the stuff I learned/taught about signal processing when I taught applied math was absitively fascinating.
23:36
Since you're doing Signal Processing, you may be interested in some AI work on timeseries and NLP.
i never knew that signal processing was so important in EE. I figured before it was only a specific biomed engg/ audio processing ting
We tend to blur the lines between strings in speech (discrete temporal sequences) and signals (continuous functions of time), and just use the results in both places regardless of the theory.
Because in the PC, everything's discrete whether we like it or not.
We throw around continuity assumptions all over the place and hope it works.
I don't really like that method of applying interdisciplinary methods, though. So I've been studying raw math for the past couple of years to learn the theory.
your right @TedShifrin if we think about them as function then it is very clear actually by construction that such sequence is in $l_{\infty}$
@Axoren very interesting stuff.
23:40
@TedShifrin I remember now what my question for you was. Do you have time to discuss the Frechet derivative?
I remember you needed the definition, let me snag it
In mathematics, the Fréchet derivative is a derivative defined on Banach spaces. Named after Maurice Fréchet, it is commonly used to generalize the derivative of a real-valued function of a single real variable to the case of a vector-valued function of multiple real variables, and to define the functional derivative used widely in the calculus of variations. Generally, it extends the idea of the derivative from real-valued functions of one real variable to functions on Banach spaces. The Fréchet derivative should be contrasted to the more general Gâteaux derivative which is a generalization of...
I want to increase my pure mathematical skill as well, since my program doesn't place as much emphasis on it as some of the other big engg schools
my department here is actually pretty strong in applied math. I will audit some classes during my grad student time while I am here.
Agh, I forgot it expands to being really big
@pingOfDoom Books. I've been reading books on bus rides and at lunch, during a small break between classes.
Any useless interval of time where you can't do anything else with it, read a textbook.
that is also what I do.
I actually finished half a book I always wanted to read as undergrad by reading it on the bus.
very productive.
In elementary school, teachers couldn't get me to read Harry Potter, ffs. Now, I'm reading textbooks by my own accord.
Such terrible persuasive skills, those teachers had.
23:43
I try to spend those useless periods finishing programming assignments that always stretch to their final hours lol
They're not useless periods if you can do homework with them.
@Axoren what do you want to know about the frechet derivativ
@syzygy I don't get why it serves the purpose of a derivative. I can't seem to justify it without using the fact that the Gateaux Derivative is equivalent to it and makes more sense.
well thats what im saying lol
Gateaux differentiability is not equivalent to it
have you seen derivatives of functions of several variables?
it is just the generalization of that concept
(jacobi matrices, say)
23:46
@syzygy Yes, they must be, at least in contexts where they're both applicable.
Shouldn't they?
Gateaux derivatives being equivalent to Frechet derivatives, not the process.
What I don't get is the process and why it works.
on wikipedia you find a counterexample
what you mean by "the process"?
Gateaux is a directional derivative
Why does the existence of the linear operator that leads that limit to be $0$ satisfy the meaning of a derivative?
i.e. along a fixed vector
in the second line you see that the definition means
that the frechet derivative is "the best linear approximation"
this is why the definition is made that way
@syzygy In the relation to the Gateaux derivative, it states that if you find the Gateaux derivative, that it is equal to the linear operator from the Frechet derivative.
yeah frechet differentiability implies gateaux derivative, and then you may compute the frechet derivative as the gateaux derivative
23:49
So the reverse isn't true, then.
precisely
Wait, hold on
Ugh, okay. So the Frechet derivative doesn't always exist.
But when it does, it's a best approximation?
yeah
I don't see that explicitly stated in the Wiki
And that wasn't expressed in class.
Equivalently, the first-order expansion holds, in Landau notation
this first order expansion says exactly that it is the best linear approximation
23:51
Oh god damn it, I think I get it.
$h$ is a vector in that expression
So, taking the limit as it approaches $0$ has to be equal to $0$ from every possible sequence converging to $0$.
Or it doesn't exist.
That's why the Gateaux derivative can exist when the Frechet derivative doesn't.
Because it only cares about the direction associated with $h$.
indeed
Question for you: Cyclic implies abelian, right?
@AaronAbraham $\vec{a} \cdot \vec{b} = \lVert{\vec{a}}\rVert \lVert{\vec{b}}\rVert \cos\theta$
the gateaux derivatives are with a FIXED direction
yes jessy
Mmm cake derivatives.
23:54
I have to digest this, I'm still not comfortable with this definition.
So if I show that a group is not abelian, I can show that it is not cyclic?
But I'm more comfortable with it. Thank you, @syzygy
axoren you should figure out
what happens in the case where V=W is the real numbers
then you get the ordinary (high school) definition of the derivative
you are welcome @Axoren
@syzygy In that case, it's damn-near trivial.
I won't be using it in that case, though.
23:56
It's interesting that you used the word "digest" right after I said cake @Axoren
@JessyCat yes
in which case do you plan to use it @Axoren ?
@syzygy thank you, my Hungarian friend
@syzygy Potentially in Matrix spaces
@s.harp then how I might I do the question?
Definitely in $n$-dimensional vector spaces
23:56
i think it is reasonable to consider the special cases with which you are familiar with
i see, so basically functions of several variables
do you know complex analysis?
I am fully comfortable with the Frechet derivative when it becomes a simple algebraic twist from a known formula. I don't know complex analysis formally.
Also, while I'm here:
0
Q: Finding all quotient groups $F/N$ up to isomorphism where $N \leqslant F \leqslant \langle H, N \rangle$.

Jessy CatSuppose $H$ is a subgroup of a group $G$, where $|H|=p$, for $p$ a prime. I need to describe, up to isomorphism, all possible quotient groups $F/N$, where $ N \trianglelefteq G$ and $N \leqslant F \leqslant \langle H, N \rangle$. Here is my attempt thus far - note that it includes a lot of subc...

I see. In any case @Axoren Gateaux derivatives are merely a generalization of partial derivatives
@Axoren: Basically Gateaux derivatives are too weak to be useful. :)

« first day (2271 days earlier)      last day (3042 days later) »