3,403 reputation
1622
bio website
location Germany
age
visits member for 2 years, 5 months
seen 3 hours ago

$1$ : It means "there exists something"

$0$ : It doesnt mean "there exists nothing", instead it means "there exist something but that is nothing"


Accept policy:

My accept policy is to accept the first completely correct answer. Acception is final and will never be changed. If no complete answer is available after a certain time, I accept partial answers.

Answering policy:

I try to answer at least two times more than the number of questions that I ask because my questions are usually two times more difficult than the questions that I am able to answer.

Voting policy:

I try to upvote all constructive answers to my questions and all nice questions/answers. I try to avoid downvoting and rather I use diplomatic ways to overcome the problems. All intentionally problematic questions or answers will be downvoted.

Respect policy:

I respect all persons who are willingly to help others. I show immediately in the comments that I dislike arrogant types.


2d
revised Does clever noise exist?
deleted 1 character in body
2d
comment Estimate large covariance matrix using few samples.
I also answered this question. Then you can simply use all the samples you have an consider sample variance estimator that I gave. It is the maximum likelihood estimator of your single parameter.
2d
comment Estimate large covariance matrix using few samples.
@nullgeppetto So in case the covariance matrix is diagonal, then it can be identified as the multiplication of $N$ marginal -uni variate- density functions. This says that for every $X_i$ we have one single Gaussian density function with some mean and variance. Mean is known from the mean vector and the covariance can be estimated as I said. Simple create a matrix of $10$ rows and $1000$ columns. Then for each column $i$ you have $10$ random samples from $X_i$ which is univariate Gaussian with some mean and variance. The problem simply reduces to variance estimation for univariate Gaussian.
2d
revised Estimate large covariance matrix using few samples.
added 13 characters in body
2d
answered Estimate large covariance matrix using few samples.
Oct
15
accepted A short question about the convexity of a function
Oct
13
comment Minimization over two lines
@YuliaV YES, definitely, no worries about it. Or I can put a bounty now but in this case I would request you to delete the answer until it is complete. Both are fine with me.
Oct
13
revised Does clever noise exist?
added 4 characters in body
Oct
13
comment Minimization over two lines
@YuliaV I can put a bounty but I am not sure if I will get an answer eventually..
Oct
12
revised Does clever noise exist?
deleted 26 characters in body
Oct
12
revised Minimization over two lines
deleted 291 characters in body
Oct
12
revised Minimization over two lines
deleted 124 characters in body
Oct
12
revised Minimization over two lines
added 41 characters in body
Oct
12
revised Minimization over two lines
added 2 characters in body
Oct
12
comment A short question about the convexity of a function
so you use binomial theorem to exchange the places of $x$ and $y$. Otherwise it would seem like $g^{-1}(1-g(1-y))=x$ and $g$ functions also change $g(n,n-k-1,\cdot)$.
Oct
12
comment Minimization over two lines
please see the edit.
Oct
12
revised Minimization over two lines
added 649 characters in body
Oct
12
comment A short question about the convexity of a function
I didnt understand two points: 1. how to reduce the problem to just showing that $g^{-1}(1-g(1-x))$ is convex? The original problem appears to be on $(x,y):g(x)+g(1-y)=1$. 2. how is $h(h(x))=x$
Oct
11
comment Exact values of error function
so you are saying that if a closed form can be found for some subset of the whole real line.
Oct
11
comment Minimization over two lines
I guess $<\theta^M$ should be $<\theta^{K-M}$. In this case $y_i=0$ can not be a solution, because it wont satisfy the constraint. If $y_i=0$ then the constaint holds iff $x_i=1$, then $(x_i,y_i)$ is neither on $l_1$ nor on $l_2$