I have used Tumblr microblogging service to organize my machine learning-related stuff for a while. You can see it here.
Although I really like the ease of tumbling, there is a big problem here: it doesn’t naturally come with searching tool and commenting system.
It happened to me that I tried to find some papers that I found a few weeks ago, but I couldn’t find it without going through all the archive. So, I think I need to come back to this blog again.
The problem is that I may just post a link to a paper without any discussion. This may distract you readers. What do you think?! Will you become unhappy if I do so?!
Approximate Dynamic Programming with Gaussian Processes
Marc Deisenroth, Jan Peter, and Carl E. Rasmussen, “Approximate Dynamic Programming with Gaussian Processes,” ACC 2008.
Similar to what we’ve done, without proofs. Formalized in a Bayesian setting (GP).
Life Goes on in Tehran
Take a look at this photoblog: Life Goes on in Tehran.
I like it because it is close to what I’ve seen there (and it is not the same as what I’ve seen on CNN!).
Evidence Contrary to the Statistical View of Boosting
In this recent JMLR paper, David Mease and Abraham Wyner experimentally show that the statistical viewpoint to boosting, which is defined as a stagewise optimization problem, does not give a sufficient interpretation for understanding the behavior of boosting algorithms (I guess they just focus on AdaBoost).
Fortunately, the paper is followed by a series of discussions by people like Andreas Buja, Yoav Freund, Robert Schapire, Jerome Friedman, Trevor Hastie, Robert Tibshirani, Peter Bickel, and other great researchers.
I have to find time and read all these papers closely. Boosting is an interesting subject for me (and probably many others). Its “automatic” and performance-dependent feature selection property deserves better understanding.
Good Writing by Marc H. Raibert
In this short piece of advice, Marc H. Raibert tells you about the way you can produce good writings. His advices are simple. He foremost suggestion is that you must believe you can write produce a good writing, and for doing so, you need to start from a [possibly] bad one and gradually improve it be several rounds of revising.
Fractional Calculus and more
I found this paper[1] about the fractional variational principle.
I didn’t know anything about fractional derivative/integration before taking a look at this paper. If you like to know more, you can check here. It gives the basic ideas.
Beside this fractional concept, there is a nice thing in the introduction of the paper which may be apparent to any physicist, but I didn’t know it:
“For every variational symmetry of the problem, there corresponds a conservation law”.
For instance, spatial invariance is equivalent to the conservation of linear momentum, and the time invariance is equivalent to the conservation of energy.
I’m wondering what does the invariance of a function w.r.t. choice of coordinate system mean (in the sense of having the same function defined on different charting of a manifold)? Some relativistic property?
[1] G. S. F. Frederico and D. F. M. Torres, “Fractional Optimal Control in the Sense of Caputo and the Fractional Noether’s Theorem,” Dec 2007, preprint.
Lectures Notes on Convex Analysis [+ Optimization]
Christian LÉONARD, “A Set of Lecture Notes on Convex Optimization with Some Applications to Probability Theory“, Lecture Notes, 2006.
Probability Inequalities and Machine Learning
Michael Steele has this course on “Probability Inequalities and Machine Learning”. This is the kind of course I’ve really liked to take. Anyway, for most of us who cannot attend the course, following his suggested reading may help somewhat.
There are some known papers/lecture notes in the reading list; such as Gabor Lugosi’s “Concentration of measure inequality” which is very readable.
There are also some other books that I hadn’t heard of it (what a shame), but seems to be nice: Pascal Massart‘s Concentration Inequalities and Model Selection.
Also in the page I found this paper by Mease and Wyner entitled “Evidence Contrary to the Statistical View of Boosting”. I definitely should read it.
well … that’s it for now!
Walter J. Freeman, “Happiness doesn’t Come in Bottles”
Have you ever felt that you are not happy from you life even if you have *objectively* successful life (e.g. a lot of money, many publications, and etc.)? And have you asked yourself what was wrong with your life?
Accidentally, I found this article “Happiness doesn’t Come in Bottles” by Walter J. Freeman. It is about happiness and its connection to the dynamical and chemical phenomenon happening in our brain. I do not want to summarize the article. Instead, I suggest you to read this short article.
I do not know if all results and suggestions in this paper are precise (the paper is a bit old and it is not written as a report of scientific discovery), but I know reading the paper was/is stimulating for myself. Maybe we geek people in the community (be in ML, CS, EE, Math, etc.) need more advice of this kind (no offense for sure!).
Scholarpedia
Have you seen Scholarpedia?
It is a wiki project on scientific subjects that are written by experts of the field and are peer reviewed by others. It means that you will know who is actually writing the most part of an article, and you know s/he is an expert.
The project is in its infancy now, so you probably cannot find the subject you like to read about, but there are several authors promising to dedicate an article to it (and the good point is that I know many of them – maybe just because they are really big figures). Currently, three main subjects are covered: Computational Intelligence, Computational Neuroscience, and Dynamical System.
This idea looks interesting to me, but I am not completely certain if it goes very well or not. The potential strength of such a Wiki project comparing to usual edited volumes is their self-sustainablity generated by people’s continual contribution. Because of rather strict conditions to start writing an article (I cannot contribute a new article unless I am really known in the field), and because of the difficulty of editing current articles (your edits should be approved by the curator of the article), I believe that the dynamics of the system is close to a damping one. I do not say these properties are bad. Actually, they bring some strength to the project (i.e. quality of experts), but they have some negative impacts too.
Whether Scholarpedia will continue to grow or just converge to a fixed point(!) is not clear to me. We may need two years to be able to predict its fate. (;
If people continue to expand this project for the next, say, 10 years, it would be a great human project. Though if it needs a constant push from its original editors, the project will not be *so* special anymore.
The other problem with the project is that it is copyrighted! I don’t like it personally, but maybe it is the best choice.
All said, I hope this project goes on well and cover other subjects of science too.