The Matrix Determinant Lemma

I had never really heard of this result, sometimes called the Matrix Determinant Lemma, but it came up in the process of answering a relatively simple question. Suppose I have an M-dimensional jointly Gaussian vector \mathbf{X} with covariance matrix A. The differential entropy of \mathbf{X} is \frac{1}{2} \log ( (2 \pi e)^M \det(A) . Suppose now I consider some rank-1 perturbation B = A + u u^T. What choice of u maximizes the differential entropy?

On the face of it, this seems intuitively easy — diagonalize A and then pick u to be the eigenvector corresponding to the smallest singular value of A. But is there an simple way to see this analytically?

Matrix Determinant Lemma. Let A be an M \times M positive definite matrix and U and V be two M \times k matrices. Then

\det(A + U V^H) = \det(A) \det(I_k + V^H A^{-1} U).

To see this, note that

\left[ \begin{array}{cc} A & -U \\ V^{H} & I \end{array} \right] = \left[ \begin{array}{cc} A & 0 \\ V^{H} & I \end{array} \right] \cdot \left[ \begin{array}{cc} I & - A^{-1} U \\ 0 & I + V^H A^{-1} U \end{array} \right],

and take determinants on both sides.

So now applying this to our problem,
\det(A + u u^T) = \det(A) ( 1 + u^T A^{-1} u )
But the right side is clearly maximized by choosing u corresponding to the largest singular value of A^{-1}, which in this case is the smallest singular value of A. Ta-da!

Teatro Luna : Living Large in a Mini Kind of Way

I saw Teatro Luna’s Living Large in a Mini Kind of Way, by Diane Rodriguez. Teatro Luna usually does pieces written by their own collective, so it was a shift for them to do someone else’s play. Living Large tells the story of Lilly, a Latina who has made it into a nice neighborhood in LA and is running for head of the neighborhood watch, but who has just lost her husband and cannot face the reality of her new lonely life — she hides bills in grocery bags in the closet and lives under the illusion that Joe has left her well-cared for. In the meantime, she tries to teach English and refinement to two domestic workers, Big Maria and Little Maria. She’s sure they have their papers (they don’t), and she strikes upon a brilliant idea to get one or both of them to move in with her. As the prospect of this increased contact looms, the comfortable deceits start to unravel. The play is a refreshing tragicomedy and strikes at the heart of the class differences and divisions in the Latino community. It’s well worth seeing, even if it is a little out of the way (The Viaduct near Western and Belmont).

Readings

A Moveable Feast (Kenneth F. Kiple) : The first half is a somewhat condensed version of the Cambridge World History of Food and covers different plants and foodstuffs from around the world. The rest of the book is about how eating habits changed over time as food exchange has diversified and now homogenized our eating habits. The only problem with the book is that it has a fair bit of apocrypha and debunked origin stories, so YMMV. I enjoyed it.

Are You My Mother? (Alison Bechdel) : Bechdel’s memoir about her relationship with her mother. It is stuffed to the brim with references to D.W. Winnicott, which can be a plus or minus depending on whether you like psychoanalysis. I thought it was engaging and worth reading, but to be honest I am not sure to whom I would recommend it. I feel like if you read the synopsis and think it sounds interesting, you will like it, and if not, you won’t.

The Learners (Chip Kidd) : This is a follow-up to The Cheese Monkeys, which I rather enjoyed. The Learners is a little leaner but still has those nerdy and fun (to me, tedious to others) asides on the art of graphic design and typography. The Milgram experiment features prominently, so if you are fascinated by that you might also like this as a piece of (sort of) historical fiction.

This Is A Bust (Ed Lin) : A novel set in New York’s Chinatown in the 1970’s and featuring Vietnam vet and alcoholic token Chinese cop Robert Chow as he struggles to turn his life around and find himself. It’s the first in a series and I will probably read the rest. Recommended for those who like detective novels.

Fisher, Neyman, and the Creation of Classical Statistics (Erich L. Lehmann) : The title says it all. It’s more about the personalities and their history than it is particularly about the math, but there’s a nice discussion at the end of Fisher’s confusing notion of fiducial probability and Neyman’s confidence intervals. I think it’s hard to put yourself back in that time period when people really didn’t have a good sense of how to think about these kind of statistical problems (maybe we still don’t have a good idea). Fisher’s work has become near-dogma now (unfortunately), but it’s interesting to see how these basic frequentist methods came about historically. Plus you get to learn more about the enigmatic “Student!” Recommended for those with an interest in the history of statistics.

IEEE PDF “Express”

So IEEE wants PDFs that appear on IEEExplore to have two properties:

  • all fonts are embedded
  • the compatibility level is 1.4

Seems simple, right? Except that their instructions for PDF Express are for those who use Adobe Distiller, which I don’t have. You’d think there would be a simple workaround, but no…

This post suggests using ps2pdf command line options, which works if all of your figures are in EPS, but not if you have PDF or JPG figures. Daniel Lemire suggests converting the PDF to PS and then back to PDF.

That didn’t really work for me — I alternately got errors saying they wanted Adobe version 5 or higher (corresponding to compatibility level 1.4) or that fonts were not embedded. I blame Mac OS. On the 10th attempt at uploading, I finally got it to work. Here’s what I did:

  1. Generate the PDF however you like (command line or TeXShop)
  2. Open the PDF in Preview, duplicate, and save a copy. This will embed the fonts but make the PDF version 1.3 or something. Say the file is called copy.pdf.
  3. In a terminal, run pdf2ps copy.pdf to generate copy.ps. This will create a PS file with the fonts embedded.
  4. Run pdf2ps14 -dEmbedAllFonts=true copy.ps to generate a new version of copy.pdf that is both 1.4 and has fonts.

This is dumb. I wasted about an hour on this idiocy and still don’t understand why it’s such a pain. It seems that on a Mac, dvips does not embed fonts properly by default, and pdflatex also cuts corners. Furthermore, it doesn’t seem like one can pass command line options (and make them default in TexShop) to automate this process.

I am sure there are better ways of doing this, but for the time being, this at least works.

A new record for the IT Transactions?

I was walking back from a seminar today and talking to Yury Makarychev and he mentioned that he and his brother Konstantin had written a paper and submitted it to the IT Transactions more than 10 years ago on a new proof of the Gács-Körner result that common information is much less than the mutual information. They submitted it, got reviews back, submitted a revised version, and then it was lost in the aether of Pareja. Now, a decade later, it is finally available to read and will appear in a future issue.

Linkage

The NIPS deadline is coming up, so I’ve been a bit harried. However, there are many cool things out there on the internet…

IIT Kanpur wants to open an office in the US to recruit faculty.

Via my father, don’t you wonder where the center of mass of a pizza slice is? This is more of an issue for those New York-style fans — in Chicago the deep dish is a little more stable.

A fascinating post from the NY Times about ephemeral islands which appear and disappear as sea levels shift.

Via BK, a musical film about coffee. It’s part of the Jazz Dance Film Fest, which promises to be my undoing, productivity-wise.

An interesting article on the Dalit movement in Maharashtra.

White pepper and honey green beans

White pepper and honey green beans

I stole this chicken recipe from Serious Eats to cook the long beans I had in my fridge. It also gave me a way to use some of the cilantro I planted in my back yard. I decided to make a veggie variant using the same marinade. This is more of a thing to riff on — I imagine it would be super tasty with brussels sprouts.

4 cloves garlic
1/3 cup cilantro, minced, stems and all (I used the leaves too since I am lazy)
2 1/2 tsp white pepper
2 1/2 Tbsp honey
2 Tbsp fish sauce (or golden mountain sauce)
1 Tbsp oil (something neutral)
1 lb (or more) Chinese long beans (or green beans) cut into 1.5″ pieces
1/2 red onion, thinly sliced (or shallots)

Grind up pepper, garlic, and cilantro in a mortar and pestle (also very important) into a paste. This will take time. Scrape out and mix in sauce and honey and oil and mix thoroughly. Marinate veggies for 10-15 minutes and then pop into an oven at 350 degrees for 15-20 minutes, mixing occasionally. Remove and carefully drain off any liquid that has sweated out, then turn on the broiler and finish them for 3-5 minutes under the broiler.

Notes: It’s really important to use white pepper. This really makes the dish. A little less honey may be good if you are not using fish sauce, as the sweetness counteracts the funk and the truly veggie version will have less of that. The liquid is a super-delicious addition to rice — I just kept in the fridge for a little zing. I’m on the fence about the oil — sesame may be good too but it might overwhelm the other flavors.