« first day (3125 days earlier)      last day (1902 days later) » 

12:03 AM
 
12:39 AM
@JossieCalderon, it's not that no inverses exist it's that vacuously, for all x not equal to zero, an inverse exists.
That is vacuously satisfied because there are no nonzero elements
A field is usually defined to have both $0$ and $1$.
I think, lol
 
1:08 AM
Hello all :) I'm currently looking for a good room to discuss machine-learning / neural networks. Could someone point me the right way?
 
 
2 hours later…
2:52 AM
Anyone know how I can reach the user @magma?
I left a comment on one of his posts with a link to an SO post, and said to comment if he's interested
He/she/they are the inventor of WildCats
Which is a closest thing on the market to BananaCats
@Anilla you stilll here?
You might want to install Orange Data Mining.
For basic enough NN's it has addons (by default) that allow you to train and try out a NN
It's not going to be as great as TensorFlow for example, but TF requires a crap load of know-how and coding
Orange: you just connect the widgets with wires
 
@BananaCats Thanks :) I'm actually building my own Neural Network from scratch just so I can learn all the ins and outs. I'd like to discuss the actual architectural/design decisions with people familiar in the field if that makes sense.
 
You might find more people like that on SO directly
stackoverflow
 
Is there a room for that?
 
Probably
go to SO
join if you have to
click upper right to chat like on MSE
 
Ohhh I see now. Thank you!! :) I'm sure I'll find something
 
3:05 AM
chat rooms
Anilla
@Anilla, do you think that NN's can be applied to diagram chasing?
Because the input is a matrix of labels which can be categorized discretely
Matrix entry (i,j) corresponds to an arrow from node i to nod j
So that is a vector as the NN input
The output is a decision of what best to do: apply functor, take an element here, etc.
 
To tell you the truth I really am not sure. I'm a medical student with an interest in math and CS... but my knowledge in that regard is quite limited
 
What kind of questions do you have about NN architecture?
 
So right now I have a small project going where I'm trying to categorize different strings. Like say I have a string: "A cat walked around in the park to find food." I'd want my NN to tell me it found related tags: "cat", "outside" and "food" for example. I would like to determine if I should use LSTM Recurrent NNs for this... or maybe I'm thinking about it all wrong
 
 
5 hours later…
8:26 AM
Hey y'all does anyone know about structure sheafs of manifolds?
 
8:36 AM
0
Q: for each point on the curve, the line segment of the tangent line from tahe curve point to the $y-$ axis has length 1.

Math geekShow that the tractrix has the following property: for each point on the curve, the line segment of the tangent line from the curve point to the $y-$ axis has length 1. My attempt:- Let the given Tractrix can be parametrised by the formula, $$c:(0,\pi/2)\to \mathbb R^2,$$ $$c(t)=\sin(t) \hat{i}...

Is my attempt correct?
 
 
3 hours later…
11:36 AM
@JuanSebastianLozano maybe, what's your question?
 
 
3 hours later…
2:23 PM
If V and W are two vector spaces, L(V, W) is isomorphic to F^n, where n is the dimension of V, what is this supposed to mean?
 
3:10 PM
Quantitatively, how erroneous would my calculations be if I considered sqrt(-1) = -1?
 
What do you mean ?
 
@Astyx Me or the guy with the isomorphism?
 
you
@FuzzyPixelz What's F ? the field ?
If so, it means $\dim V \times \dim W = n$
 
3:26 PM
oh, I meant like, I know `i = sqrt(-1)` and a complex number `z = a + ib`, where `a` and `b` are real numbers. But say I considered i to be some other real value, not imaginary. I'm assuming `i = -1`, by what measure will z be affected. How do you measure that? The difference between `z1 = a + ib` and `z2 = a - b + 0i`.
It feels like I'd just be studying `Re(z2 - z1) = b`which is not useful. Is there anything interesting about the plane on which the (1+i)/|1+i| unit lies? I think it's on the diagonal of a box, but like in this idea .. I'm like thinking `i=-1`, so `0/0` is the diagonal. I
@Astyx or leave it, that doesn't make much sense actually :I I'm thinking a lot.
i'm feeling 0/0 means that diagonal plane is the only plane on which a direct shadow from neither real plane or imaginary plane can fall. (my language skills need improvement probably)
no, wait, that is wrong. I've forgotten what cases I was imagining :/
sorry to waste your time :I :I :I
 
4:17 PM
The main issue is that $(-1)^2 = 1$
So yeah, it doesn't make much sense
 
4:33 PM
My bad @Astyx there is no F it's just W... I was trying to avoid french notation and this happened haha
I know the definition but isn't there any insight ?
Like when we say that every matrix is associated with a linear transformation
 
Then it means $\dim V = n$
If you take the coordinate functions $e^*_1,\dots, e^*_m$ on a basis $e_1,\dots, e_m$ of $W$
Then any morphism $\phi \in L(V,W)$ is characterised by the $e^*_k\circ \phi$
which are elements of $L(V, \Bbb R) = V^*$
This space is called the dual of $V$ and has the specificity of being isomorphic to $V$ in finite dimensions
(the coordinate functions of a basis of $V$ form a basis of $V^*$, which gives you a direct isomorphism)
So in this case, any elements of $L(V,W)$ is uniquely defined by $n$ elements of $V^*$, ie an element of $(V^*)^n$
This is isomorphic to $V^n$ by what I stated above
Meaning that $L(V, W)$ is isomorphic to $W^n$
More generally, in finite dimensions, spaces are isomorphic iff they have the same dimension
@FuzzyPixelz
 
4:51 PM
Does anyone here know about Mackey's imprimitivity theorem in abstract harmonic analysis?
That is a lovely picture @AlessandroCodenotti.
 
5:40 PM
@MatheinBoulomenos You are correct.
 
5:56 PM
What's a good structure preserving map from the lattice of the natural numbers in the plane to the unit square?
does one exist?
 
@AlessandroCodenotti it's a soft one, but pretty much I just want to know what the structure sheaf tells you geometrically, like why is it useful? I am thinking here of manifolds, it seems clear to me why it matters, say, in a scheme
 
What structure do you want to preserve exactly? @Ultradark
 
I just want the natural numbers to all be in the unit square. I don't care if the grid has to be warped in order for this to happen @PaulPlummer
 
Maybe if someone wants to get this question a go:

https://math.stackexchange.com/questions/3122924/comparing-sigma-algebras-with-topologies
 
so not preserving any structure?
 
6:06 PM
@MatheinBoulomenos And also, the composition in the case that $R$ is non-commutative is not well-defined.
 
@PaulPlummer well I do want the ordered pairs of natural numbers to be next to each other and not completely randomly scrambled
for example $(3,2)$ should remain one edge away from $(4,2)$
but I'm not sure if this property is "distance preserving"
 
Perhaps think about it as the complex plane, and do some thing like $1/z$ (probably scaled and shifted a bit depending on what you want)
 
I'm going to try to do it without complex analysis
 
6:22 PM
It is hardly complex analysis, it is circle inversion
complex numbers are a convenient way to do it though
 
okay thanks I got it
 
6:49 PM
now I want to map all the even numbers of the lattice of natural numbers in the unit square to the vertical line $x=3/4$
 
Anonymous
@AlessandroCodenotti So, you can view all conversations bookmarked by you on this tab. (Late reponse, I know. :P)
 
vzn
@Anilla hi would also like to see more activity on ML in SE chat rooms. have had intermittent discussions in these rooms, there are a few scattered users/ experts, can introduce you to some, have worked in area myself, somewhat more with GAs etc chat.stackexchange.com/rooms/71/the-h-bar chat.stackexchange.com/rooms/43371/the-singularity re your question look into natural language processing
 
and all the odd numbers to $x=1/4$
 
 
2 hours later…
8:35 PM
@vzn Thank you! :)
 
9:07 PM
1
Q: Formal group law and Koenigs function conjecture !?

mickLet $f(x,y)$ be a symmetric real function and a formal group law $$G(x + y) = f(G(x),G(y)). $$ Consider the equation $$ h(2x) = f(h(x),h(x)) = A(h(x)). $$ This equation has many solutions. Compute a solution to that equation with the fixed point at $0$ and its Koenigs function, and call the ...

Any ideas ?
This is important to me
Should I crosspost to mathoverflow ? I assume there is a simple solution , so I assume better not.
 
9:36 PM
@Blue that's great, thanks!
 
hi, italic @Alessandro
 
hi ted
 
heya @JoeShmo.
Back to the safety of snow on the east coast?
 
yes. thoroughly freezing
 
9:39 PM
Well, southern CA residents consider the recent temperatures in the 40s to be freezing.
 
i know, they would all complain. cry me a river..
the snow is largely gone now
 
9:54 PM
what curve goes through the most primes less than $100$
 
Hi chat
 
y=0 or y=x or x=0
hi astyx
I'd like to plot all the coordinates of the form $(p,p)$ in an $n \times n$ lattice
see what the pattern looks like
 
what do you mean plot? $p$ is prime?
 
yeah $p$ is prime. What I mean by plot is "highlight" all the points in the lattice of the form $(p,p)$
and leave the other points the same color
 
10:09 PM
right. the question wouldn't make sense if $p$ wasn't prime..
i bet there's plenty of codez online that accomplish just that
 
I can only find spiral patterns and such. Nobody wants to stick with a regular grid it seems
 
well, ok
find all primes $< 100$. Plot them on the plane
The sieve of Eratosthenes is what you're after
the spiral youre talking about is due to Ulam
presumably
 
10:28 PM
I'm going to plot it and post it here
 
nice! looking forward
 
10:49 PM
 
Looks cool, what about up to 1000?
 
give me another 20 min
 
Hahaha, okay
 
Wait, how is that of the form $(p, p)$? Shouldn't it resemble more of a diagonal line? (Still looks cool btw).
 
Hi chat.
 
10:58 PM
Oh you're doing $(p, q)$ where $p$, and $q$ are primes.
 
Let $a,b,c \in \mathbb{C}$.
And let $j \in \mathbb{C}$ such that $j^2+j+1=0$. Show that $a^2+b^2+c^2=ab+ac+bc$ iff $j$ and $j^2$ are the solutions of $az^2+bz+c=0$.
I'm trying to show this nowadays.
 
11:34 PM
Is is true that $\int_a^b \lim\limits_{n \to \infty} f_n(x)=\lim\limits_{n \to \infty} \int_a^b f_n(x)$ even if $f_n$ does not converge uniformly?
 
yes
 
@Ultradark I can't think of an example.
 
It can be easily seen that if $j$ is a solution to this equation, then so $j^2$ because $j^3=1$ and there is a symmetry.
 
@user330477 $f_n(x)=x^n$
 
11:49 PM
@Ultradark But then you should really have $\abs{x}<1$. Does something like $f_n(x)=\frac{2x}{n}+\frac{1}{n^2}$ work?
 
@user330477 Let $f_n$ be equal to $\frac{1}{\frac{1}{n}-\frac{1}{n+1}}$ on $(\frac{1}{n+1},\frac{1}{n})$ and $0$ everywhere else on $[0,1]$.
 
$\dfrac{1}{\frac1n - \frac1{n+1}} = \dfrac{1}{\frac{1}{n(n+1)}} = n(n+1)$
 
@MatheinBoulomenos Thank you, I am checking your example. Is the answer provided by me correct?
 
your $f_n$ do seem to converge uniformly on closed intervals
 

« first day (3125 days earlier)      last day (1902 days later) »