13,925 reputation
32777
bio website suvrit.de
location Internet
age
visits member for 4 years, 5 months
seen 1 hour ago

Researcher in Optimization, Machine Learning, etc.

Hobbyist in a small fraction of the world of Inequalities, Matrix Analysis, Noncommutative polynomials.

I'm here mostly to learn old and new mathematics!


1d
comment Projected Alternating Minimization
No, it does not. Have a look at the explicit example in Fig. 1 here: www3.interscience.wiley.com/cgi-bin/fulltext/117874500/PDFSTART
1d
comment Does this symmetrization operator have a name? Any theory?
I think this may be called a "polarization operator" --- see e.g., page 6 of Gurvits' paper: arxiv.org/abs/math/0510452v3
1d
reviewed Close Difference between Principal Component Analysis(PCA) and Singular Value Decomposition(SVD)?
2d
comment Is this parametric inequality true?
@TerryTao: Curiously, checking $\log(\cosh(x))/x^2$ is decreasing also showed up here, though again not without calculus: mathoverflow.net/questions/40334/…
Jan
27
comment Finding sparsest solution of a linear system
Search for: Iterative Hard Thresholding --- that works as a reasonable heuristic (with some guarantees under compressed sensing style assumptions); in any case, something that you can code up in 3 lines of Matlab.
Jan
26
reviewed Close the triangle inequality for shortest paths of graphs
Jan
25
reviewed Close Ky Fan norms and nuclear norm
Jan
23
comment Subadditivity of the square root for matrices
@GottfriedHelms: Yes, for commuting matrices, the question will boil down to $|a^r - b^r| \le |a-b|^r$ for $a,b \ge 0$.
Jan
21
reviewed Close Why Laplacian Matrix need normalization and how come the sqrt of Degree Matrix?
Jan
21
reviewed Close Analytic minimization of linear algebraic expression with nonlinear constraints
Jan
21
reviewed Close Vanishing Points using Homogeneous Coordinates
Jan
20
comment Computation Complexity for Golden Section method
The computational complexity here is not referring to arithmetic complexity but rather oracle complexity, so that the golden section method, which converges to an $\epsilon$-accurate solution at a linear rate (also known as geometric / exponential convergence in numerical literature), will make $O(\log (1 / \epsilon) )$ calls to the oracle (to compute the value of $f(x)$) --- thus @Shamisen's pointer is sufficient.
Jan
17
comment Subadditivity of the square root for matrices
@MateuszWasilewski: Lower bounds on $\sigma_n$ are just too useful to be easily had :-) --- even the majorization result that I cited was quite nontrivial and took a few years of effort to get proved!
Jan
17
revised Subadditivity of the square root for matrices
added explicit example and more detailed citation to result that holds.
Jan
16
reviewed Close Sum of Log Normal random variables
Jan
16
reviewed Close QR decomposition of matrix
Jan
16
answered Subadditivity of the square root for matrices
Jan
16
revised Eigenvalues of product of two symmetric matrices
changed majorization to weak maj
Jan
9
awarded  reference-request
Nov
29
awarded  Enlightened