dr Łukasz Dębowski

 me

My research interests revolve around probability, language, information, and learning, and I try to weave together very different threads I know.

Until the end of 2009, I am working with the Information Theoretic Learning group in the CWI. I am on leave from the IPI PAN, where I have worked with the Statistical Analysis and Modeling group and co-worked with Linguistic Engineering. Seeking for big AND purposeful AND feasible intellectual adventures, I have first studied at the Faculty of Physics, UW. Later I also visited the UFAL, the Santa Fe Institute, and the CSE UNSW. Many interesting people showed me strikingly different ideas what is worth doing in alpha and beta sciences, in engineering, and in general ;). I slowly realize what I should and can do best myself, thoroughly exploring various options on my way, with a great speed-up provided by the Internet, the knowledge accumulated so far, the growing circle of my friends, and a few moving gaps in my education and memory.—That is what I do.

Instead of a research statement, browse the preprints below.

I seem good in paraphrasing and modyfying abstract mathematical problems so that they acquire both empirically interesting meanings and simple solutions. This seems rare and may come in part from ruminating broad non-mathematical experience. If you disagree, tell me. I wonder how I work. I like Lakatos' Proofs and Refutations.

I am expected to have an assistant once I am back to IPI PAN.

If you think that you can be my assistant, write me an email.

You can mail me either to ldebowsk@ipipan.waw.pl or to debowski@cwi.nl. For other contact data, see here.

Recent preprints and publications
(a) Computable Bayesian Compression for Uniformly Discretizable Statistical Models. pdf (Appeared in ALT'09; this version contains two corrections.)
For an effectively identifiable parameter, Bayesian compression is the best enumerable compression if and only if the parameter is Martin-Löf random. Cf. also (d).
(b) On the Vocabulary of Grammar-Based Codes and the Logical Consistency of Texts. arXiv (Under review in IEEE Trans. on Inf. Theory.)
The summary of a new explanation of Zipf's law in linguistics, under puzzlingly general conditions: The texts contain so many distinct words because they consistently describe so many distinct facts.
(c) Variable Length Coding of Two-Sided Asymptotically Mean Stationary Measures. pdf (Under review in J. of Theor. Probab.)
A few technical results that concern stationarized variable length coding of stationary measures, useful for the manuscript (c).
(d) A general definition of conditional information and its application to ergodic decomposition. pdf (Appeared in Stat. & Prob. Lett.)
Some other little results useful for the manuscript (c). Unexpectedly, they resonate with (a).
(e) Valence extraction using EM selection and co-occurrence matrices. arXiv (To appear in Lang. Res. & Eval.)
I wanted to understand what the verb valence frame can be from a statistical AND algebraic point of view.
(For the citation details, check the publications' link below.)

Serious stuff
Nerd's fun
Łukasz Dębowski, 08.10.2009