Every now and then, Vapnik lets drops of wit fall on the pages, and the reader that can spot them is able to savour the wisdom of a great mind.

I like this one, which was “buried” in a footnote on page 13:

“Of course it is very interesting to know how humans can learn. However, this is not necessarily the best way for creating an artificial learning machine. It has been noted that the study of birds flying was not very useful in constructing the airplane.”

Interesting thought, and nice counterexample. =)

Posted from Providence, Rhode Island, United States.

• Luigi Pirandello — Le Novelle
• Italo Calvino — Lezioni Americane
• Tommaso Labranca — Neoproletariato
• Vladimir N. Vapnik — The Nature of Statistical Learning Theory

In valigia, anche le bozze di tre articoli: uno in via di pubblicazione, uno in divenire, uno da finalizzare. In testa, un’idea per un nuovo articolo.

Posted from Padova, Veneto, Italy.

Anyway, the following is a passage from the Preface to the First Edition of Vladimir N. Vapnik’s The Nature of Statistical Learning Theory, 2^nd Ed.

[…] during the last few years at different computer science conferences, I heard reiteration of the following claim:

    Complex theories do not work; simple algorithms do.

[…] I would like to demonstrate that in this area of science^[1] a good old principle is valid:

    Nothing is more practical than a good theory.

Vapnik does not say (at least in this Preface) whether a “good” theory could also be simple or actually have to be complex, nor whether the complexity of the theory has any significance on its being “good”.

Still, I like this passage. It is bold, effective, and probably true, especially in the part regarding computer scientists’ point of view.

1. ^ statistical inference.

Posted from Providence, Rhode Island, United States.

“Machine Learning has become a very mathematical and statistical-based research area yet the theoretical computer science community hasn’t played the role in this area that we could have.”

From a blog post by Lance Fortnow

Posted from Providence, Rhode Island, United States.

Ora caffè da Loui’s, con pancake e coppa di frutta. Yay.

Confesso che alle 5.27 pensavo ci fosse luce fuori, invece è ancora tutto buio.

Posted from Providence, Rhode Island, United States.

The fact is that those reviews refer to the first printing of the second edition of the book, which was indeed flawed with tons of typos, as shown by the 15-pages-long errata. If you ever think of getting the book, be sure to get the “Updated Printing” version. There are still some typos, but not more than in any scientific book (I retain the right to revise this statement as I go on reading the book). I still have no opinion about the index, whether it was improved or not, since I am using it as a textbook and not as a research reference.

As far as the contents of the book are concerned, I am not ready yet to do a full review and I doubt I’ll ever be, but I have to say that the Authors make lots of references to the many examples they present, which is, in my opinion, not such a good things to do, because you can find sentences like:”The problems like the one presented in Example X.Y.Z. are called foobar“. Now, extrapolating a class of problems from a single example is, at least for me, a very difficult inference task (:)), so I would have preferred more formal definitions and a less heavy use of examples to explain fundamental concepts. Examples should help in understanding previously formally defined concept, not be used as implicit definitions. I am also quite puzzled by some notational choices: why on Earth did the Authors choose to denote the set of reals number as $R$ and not as the more classic $\mathbb{R}$ (yes, I should install a LaTeX WordPress plugin). The style is pretty dry, but I did not expect anything different, so I don’t consider this an issue.
Something else bothers me instead: the fact that foot notes are not printed at the foot of the page they appear in, as the name should suggest. They are instead collected at the end of each chapter, which makes checking them quite annoying. Luckily there aren’t many footnotes.

I have to note that the Authors claim, in the Preface to the Second Edition, that Volume II of the book is expected in 2010, with topics like nonparametric methods, bootstrap and Markov Chain Monte Carlo. I don’t see any sign of the existence of this book anywhere, but I hope the Authors will do a better job in having typos fixed before printing the book.

This review may and should be expanded. =)

I have another bad tale involving this book but the main “villain” of the novel is actually Amazon and not the book nor its Authors. I bought a first copy of the book from Amazon (I mean, directly from Amazon, not from a 3rd party seller listed on Amazon) and it was one of the last two left in stock (the book is currently out of stock on Amazon, but should be available shortly). The description clearly said that the volume is hardcover, but the copy I got was softcover. What is much worse, the pages inside were not printed: they were non-page-aligned scans/xeroxs of a genuine copy of the book. Clearly I was not satisfied given that I paid around $60 for my copy, therefore I filed for return and full refund. Meanwhile, I ordered another copy of the book, this time from a 3rd part seller listed on Amazon. This time I got a genuine print. I still have to send my bad copy back, but I’ll update this post when I get the refund (or if I do not get it).

Posted from Providence, Rhode Island, United States.

Personalmente la trovo fuffa, nel senso che se hai bisogno di un corso per imparare a gestire le relazioni con il tuo advisor, una persona che, in un modo o nell’altro, dovresti esserti scelto, allora forse dovresti avere bisogno anche di un corso per imparare a fare ricerca e magari per relazionarti con il mondo in generale. Mi sembra, insomma, un “ti porto per manina come se fossimo all’asilo, anche se hai un’età e un ruolo in cui dovresti essere in grado di tirare fuori le palle e far riconoscere quanto vali”.

Faccio insomma mia la convinzione di Roberto Vecchioni:

“I ragazzi devono faticare. Basta accarezzarli, fategli il culo fin da piccoli.”

Aggiungo, che personalmente non direi mai di no al mio advisor, qualunque progetto extra mi chiedesse di fare, e non per lecchinaggio, ma nella convinzione che se me lo chiede vuol dire che pensa che possa venirne fuori qualcosa di buono per me.

Piccola osservazione finale: io, e come me tutti i Ph.D. student qui a Brown, non ho nemmeno un vero e proprio contratto con nessuno. Non sono un dipendente di Brown University e neppure del mio advisor, tuttavia ricevo uno stipendio pagato con i grant ottenuti da lui ottenuti.

Posted from Providence, Rhode Island, United States.

Una citazione per tutte da “The Fractal Geometry of Nature”:

“Clouds are not spheres, mountains are not cones, coastlines are not circles, and bark is not smooth, nor does lightning travel in a straight line.”

(non so se l’ultima sia davvero una “conseguenza” dei frattali, io pensavo fosse una conseguenza della gravità, ma magari le cose sono collegate).

Alcuni dei volumi di dispense del mio corso di Analisi 1, il mio primo esame all’università, autori O. Stefani e A. Zanardo, avevano in copertina dei bellissimi frattali a tema naturalistico, in particolare ne ricordo uno con i pesci (in realtà non sono certissimo fossero frattali, erano piuttosto partizioni ricorsive, ma magari rientra nella definizione).

Dal blog di Lance Fortnow:

“One of the popular screen savers on the Sun computers in the late 80′s was either zooming in on the Mandelbrot set to reveal the same set inside. One could spend hours watching–this and Tetris wasted many grad student hours in those years before Facebook and Twitter”.

=)

Posted from Providence, Rhode Island, United States.