I have spent some time recently studying matrix functions, both from theoretical and computational perspective. There is a nice book by Nick J. Higham on functions of matrices, which I highly recommend to interested reader and which provides a thorough overview of current theoretical results on matrix functions and several efficient numerical methods for computing them. Another well written text is by Rajendra Bhatia on matrix analysis (graduate texts in mathematics), which includes topics such as the theory of majorization, variational principles for eigenvalues, operator monotone and convex functions, matrix inequalities and perturbation of matrix functions. Bhatia's book is more functional analytic in spirit, whereas Higham's book focuses more on numerical linear algebra.

Below you will find a report that I produced and which contains a few interesting (some are elementary) proofs and implementations of algorithms. Interested reader should check the literature above to be able to follow the text.

Abstract: Fast growth in the amount of data emerging from studies across various scientific disciplines and engineering requires alternative approaches to understand large and complex data sets in order to turn data into useful knowledge. Topological methods are making an increasing contribution in revealing patterns and shapes of high-dimensional data sets. Ideas, such as studying the shapes in a coordinate free ways, compressed representations and invariance to data deformations are important when one is dealing with large data sets. In this talk we consider which key concepts make topological methods appropriate for data analysis and survey some machine learning techniques proposed in the literature, which exploit them. We illustrate their utility with examples from computational biology, text classification and data visualization.

In May I participated at the first BioDay meeting organized by Hekovnik in Ljubljana, Slovenia. The aim of the BioDay events is the exchange of ideas, knowledge and fostering collaboration and networking between life scientists, computer scientists, bioinformaticians, mathematicians and physicists.

The first event focused on recent trends in bioinformatics, specifically on experimental methods in systems biology (by Spela Baebler, PhD) and biomedical data fusion. I presented the latter topic and discussed how heterogeneous data sources in biology can be collectively mined by data fusion. The video of the event is available at video.hekovnik.com/bioday_trendi_v_bioinformatiki (in Slovene). Enjoy!

The goal of the challenge was to assess the performance of information extraction systems to extract a gene regulation network of a specific cellular function in Bacillus Subutilis. This function was sporulation and is related to the adaptation of bacteria to scarce resource conditions. The automatic reconstruction of gene regulation networks is of great importance in biology, because it furthers the understanding of cellular regulation systems.

We were provided a manually curated annotation of the training corpus including entities, events and relations with gene interactions. Also, the regulation network that can be reconstructed with interactions mentioned in sentences of training data was provided (picture on the right). The task required to estimate gene regulation network from test data by specifying a directed graph where vertices represent genes, and arcs represent interactions between genes extracted from the text. The arcs were labeled with an interaction type (e.g., inhibition, activation, binding, transcription).

We hope to describe our approach using conditional random fields and rules in a paper but the details are not public yet (stay tuned).

Some of the ACM XRDS Editors are participating these days in a meeting to discuss the magazine's future direction in print and online. We will do our best to further promote the XRDS, enhance its departments, improve web presence, build a community of readers, and provide high quality content from various CS disciplines.

See the current ACM XRDS issue on Information and Communication Technologies and Development (ICTD). The Spring issue is coming very soon! It will be on Scientific Computing. And, the Summer issue will be on Computer Science and Creativity.

If you are interested in submitting an article or have some crazy good ideas, please share them with us (contact the editorial staff). Or, share with your colleagues any interesting columns/featured articles you read in ACM XRDS.

XRDS is the ACM's flagship magazine for students, established in 1994. It is published quarterly and invites submissions of high quality articles of interest to computer science students (from Editorial Calendar).

In first week of December U of Ljubljana celebrates traditional "Week of University" (Why?) during which numerous invited lectures, presentations and award ceremonies are organized.

This year, I was awarded the faculty award for best students and Preseren Award of U of Ljubljana for thesis "A Matrix Factorization Approach for Inference of Prediction Models from Heterogeneous Data Sources" (slo: univerzitetna Prešernova nagrada za delo "Pristop matrične faktorizacije za gradnjo napovednih modelov iz heterogenih podatkovnih virov"). I would like to thank my supervisor and mentor Prof. dr. Blaž Zupan for encouragement and advice he provides throughout my time as his student. I am lucky to have a supervisor who cares so much about my work and responds to my questions promptly. I could not have won the award without his support and mentoring.

My work here will be about network integration for disease classification, specifically inferring prediction models from heterogenous data sources through matrix factorization. More about it in the next days. For now you can check the Interactive map of the Diseasome (Below is an image showing a part of diseasome. Interested reader is referred to Barabasi's paper The Human Disease Network.) linked from the NYTimes article Redefining Disease, Genes and All.

Modeling the regulatory diversity of human cancers (S. Nelander)

Tissue specific modeling of functional genomics data: from networks to understanding human disease (O. Troyanskaya)

An evaluation of methods for the modeling of transcription factor sequence specificity (M. T. Weirauch)

SEEK and find: data management for systems biology projects (O. Krebs)

Excerbt: next-generation knowledge extraction and hypothesis generation from massive amounts of biomedical literature (B. Wachinger)

Combining multiple biological domains using patient network fusion (B. Wang)

Combining many interaction networks to predict gene function and analyze gene lists (Q. Morris)

Assembling global maps of cellular function through integrative analysis of physical and genetic networks (R. K. Srivas)

iCAVE: immersive 3d visualization of biomolecular interaction network (Z. Gumus)

Systems-level insights from the global yeast genetic interaction network (C. Myers)

Monopoly systems edition: advance to GO collect $200 (T. Idekar) (*actually about NeXO, a network extracted ontology and functional enrichment)

Genome-scale metabolic models: a bridge between bioinformatics and systems biology (J. Nielsen)

The organizers came up with a nice social program, parts of it is depicted on images below. At opening ceremony Tanja Tagaq, an Inuit woman, performed a unique style of traditional throat singing, Amanda and Rasmus from Sweden made performance at first poster session, Serena Ryder entertained us at conference reception dinner. Shonen Knife, a Japanese punk band that opened Nirvana, played at second poster session at Hart House.

I attended workshops on Designing experiments using state of the art Bayesian global parameter search methodology (M. Goldstein), Introduction to the statistical inference or regulatory networks (F. Emmert-Streib), Imaging flow cytometry: a new view on systems biology (R. DeMarco). In addition to parallel sessions I also enjoyed special lectures and plenary sessions. A few of them are: Reading and writing omes (G. Church), Towards unification of genetic and hierarchy models of tumor heterogeneity (J. Dick), Interactome networks and human disease (M. Vidal), The genetics of individuals (B. Lehner), Synthetic genetic interaction analysis by high-throughput imaging to map cellular networks, Unraveling principles of gene regulation using thousands of designed promotor sequences (E. Segal), Systems biology applications of imaging flow cytometry (T. Galitski).

So what have I been up to in recent weeks here at Toronto? Highlights include my first ride with famous American yellow school bus to a reception at ICSB12 conference, some sightseeing in Toronto city and a trip to Niagara Falls.

Besides, I have finished with data analysis of real-time yeast S. cerevisiae microscopy screens, an idea about it can be captured here. I am now starting with time series analysis and will probably have time to work on integration of phenomics data with genetic interaction and protein interaction data.

Recently a quantum optics research group here at UofT demonstrated a violation of Heisenberg's uncertainty principle and I was really excited about their work. "The quantum world is still full of uncertainty, but at least our attempts to look at it don't have to add as much uncertainty as we used to think!" ... and an easy reading to motivate you to learn more.

I have also come upon a nice real-world (I do not like this term) implementation of an argument based machine learning offered through classification module in CellProfiler Analyst package, participated in a discussion about Gaussian processes (intro, notes) at ccbr-stats meeting and much more. The Lab organized a farewell lunch for summer students only two weeks after my arrival to Toronto, as here and in US classes have already begun (after the Labour Day), I considered it as a welcome event :)

Below are images of Toronto CN Tower, Niagara Falls as seen from Skylon Tower and squirrels at UofT campus (Yes, one cannot miss numerous squirrels playing in parks at campus. A careful look should reveal four of them.), respectively.

My work will be mostly data analysis of S. cerevisiae screens by employing various statistical and machine learning methods to gain new knowledge about identification of yeast mutant strains with non-WT phenotype. Possibly I will also work on time-series analysis of actin patches in yeast cells to differentiate them. First impressions are great, I have already met some great people and am looking forward to meet some at the International Conference on Systems Biology (ICSB12), which is held in Toronto in the next week and have a fortunate opportunity to attend.

Professor of Computer Science Dick Lipton from Georgia Tech has a few days ago posted an interesting post on his blog about a new way of solving systems of linear equations over prime fields. Solving linear systems of equations is a very well explored area and a new approach came as a pleasant surprise to me. The author of the proposed method is Prasad Raghavendra from Berekeley. He wrote a draft paper with the algorithm and the analysis in which he proves the soundness and completeness of the randomized algorithm he had proposed. The proof is not very complicated, you need to be familiar with some probabilistic bounds, specifically a generalization of the Bernstein inequality, the Hoeffding inequality, and foundations of finite fields.

Solving linear systems of equations is indeed an ancient problem and a new approach is therefore warmly welcomed. As early as 200 BC the Chinese has devised a clever method for solving 2-by-2 systems and there is evidence that they had developed a method essentially equivalent to Gaussian elimination (2000 years before Gauss came up with it!). Then Gabriel Cramer has in 1750 published a famous and simple O(n^3) rule for solving system of linear equations by computing matrix determinants and cofactors. Later in 1810, Carl Friedrich Gauss became interested in discovering the orbit of Pallas, the second-largest asteroid of the solar system. His work led him to a system of six linear equations in six unknowns. Gauss invented the method of Gaussian elimination which is still used today. The method consists of performing row-operations on the coefficient matrix to obtain an equivalent system of equations whose coefficient matrix is upper-triangular. This means that the last equation will involve only one unknown and can be easily solved. Substituting that solution into the second-to-last equation, one can then solve for another unknown. Gauss' work was continued by Wilhelm Jordan who convert a given system of linear equations to a triangular system which can be reduced to a diagonal one and then solved directly (Gauss-Jordan method) -- LU decomposition. We now have faster methods for general systems based on discoveries of Volker Strassen and almost linear time algorithms which require systems to have special structure.

Most readers probably know methods mentioned above very well, so let us do not spend any time discussing them. So, what is the trick in the new method? Prasad starts with the set of random vectors, V0. We expect that about half of them will satisfy the first linear equation. He then throws away all vectors that do not satisfy the first equation and call the remaining set S1. The idea is successively applied to further equations until we reach the set Sm. Vectors in Sm will then satisfy all linear equations and represent the solutions to the system. The problem with this winnowing-down process is that with high probability the set Si winnows down to zero before all equations in the system are satisfied. An exception is if we start with a set V0 of exponential size, but this is obviously too much.

Prasad came up with a brilliant answer that indeed we do not need a large initial V0 (he proved it can be of linear order in the number of variables) and still have a high probability of getting a solution to the complete system of linear equations if the system is feasible. The trick is to use an amplifier and a property of finite fields. Remember, the method is for systems of equations over finite field and the exploited property limits its usage. Over finite field with characteristics p, the trick is to form Vi (being of the same size as V0) as sums of random p+1 vectors from Si. Namely, summing p+1 solutions to the first p equations will give in finite field a (new) solution to the first p equations and possibly solve also (p+1)-th solution. See the draft paper for the details.

The new algorithm is easy to implement. Recently I am becoming quite a fan of Clojure, so I decided to implement the algorithm in this popular functional programming language that runs on JVM. I have always been very fond of functional programming languages (examples include Haskell, OCaml, F#, Lisp, ML, Erlang, Scheme etc.) and I am glad functional paradigm is gaining popularity also in "real-world" problems (I do not like this term, but ...). Functional programming is not more difficult than object oriented programming, it is just different. It is a style of programming that emphasizes first-class functions that are pure and is inspired by ideas from lambda calculus. I acknowledge that Matlab would be the most appropriate for this problem and imperative language implementation more understandable but it is more fun, isn't it :) ?

(ns clj_concurrency.ls-solver)
(declare n m q N)
(defn random-ints [X] (conj X (repeatedly n #(rand-int q))))
(defn mod-reduce [X] (map #(map (fn [x] (mod x q)) %1) X))
(defn satisfy? [Ax Sx bx]
(= (mod (apply + (map * Ax Sx)) q) bx))
(defn select [S Ai bi]
(def T #{})
(doseq [Sx S :when (satisfy? Ai Sx bi)]
(def T (conj T Sx)))
(into #{} T))
(defn recombine [T]
(def R #{})
(dotimes [_ N]
(let [Q (take (+ q 1) (shuffle T))]
(def R (conj R (reduce #(map + %1 %2) Q)))))
(into #{} R))
(defn solve [A b]
(def S (apply concat (map random-ints (repeat N []))))
(dotimes [i m]
(let [T (select S (nth A i) (nth b i))]
(when (empty? T)
(throw (IllegalArgumentException. "System infeasible, T.")))
(def S (recombine T))))
(if (empty? S)
(throw (IllegalArgumentException. "System infeasible, S.")))
(do
(println "--------")
(println "Solution" (distinct (mod-reduce S)))))
; toy example 1
(def n 3) ;number of variables
(def m 3) ;number of equations
(def q 3) ;finite field characteristics
(def N (* 30 n)) ;sampling size
(solve [[2 1 1] [1 1 1] [1 2 1]] [1 0 0]) ;--> (1 0 2)
; toy example 2
(def q 5)
(def m 3)
(def n 3)
(def N (* 30 n))
(solve [[1 1 1] [2 3 2] [1 3 4]] [1 4 4]) ;--> (1 2 3)

With that Thesis I completed a four year interdisciplinary university program consisting of eight semesters of lectures and one year of Diploma thesis work in less than four years. Diploma leads to the degree of University dipl.ing. of Computer Science and Mathematics. I graduated summa cum laude with the average grade being 10.0 out of 10.0.

My brother Slavko took some pictures and shot the defense. The movie and pictures are available at zitnik.si site.

Abstract

Today we are witnessing rapid growth of data both in quantity and variety in all areas of human endeavour. Integrative treatment of these sources of information is a major challenge. We propose a new computation framework for inference of prediction models based on symmetric penalized matrix tri-factorization and intermediate strategy for data integration. Major advantages of the approach are an elegant mathematical formulation of the problem, an integration of any kind of data that can be expressed in matrix form, and high predictive accuracy.

We tested the effectiveness of the proposed framework on predicting gene annotations of social amoebae D. dictyostelium. The developed model integrates gene expressions, protein-protein interactions and known gene annotations. The model achieves higher accuracy than standard techniques of early and late integration, which combine inputs and predictions, respectively, and have in the past been favourably reported for their accuracy.

With the proposed approach we have also predicted that there is a set of genes of D. dictyostelium that may have a role in bacterial resistance and which were previously not associated with this function. Until now, only a handful of genes were known to participate in related bacterial recognition pathways. Expanding the list of such genes is crucial in the studies of mechanisms for bacterial resistance and can contribute to the research in development of alternative antibacterial therapy. Our predictions were experimentally confirmed in wet-lab experiments at the collaborating institution (Baylor College of Medicine, Houston, USA).

Note: Complete Thesis will be available for download after some aspects of the proposed approach will be presented in the scientific journal.

I am looking forward to starting my PhD studies in the autumn. Until then ... many new interesting projects ...

I have been following the Cryptography and Theory of Coding 2 course this semester (summer 12).

During the course I have been working on the project about Commitment schemes. Please find below attached the final report produced and a short presentation. Additionally, some homework solutions are attached as well. The language is Slovene, but it is mostly pure math with proofs, so it should not be too difficult to capture the main idea.

Please note these are reports and have not been subject to the usual scrutiny of published papers.

XRDS, Crossroads is the flagship academic magazine for student members of the ACM. It is published quarterly, both in print and online for ACM student members. The publication was previously named Crossroads and has been running since 1994, edited and maintained voluntarily by students. The magazine is distributed to tens of thousands of students worldwide.

Issues focus on Computer Science and include various articles from highlights in new and important research to roundtable discussions, interviews and introductory overviews. In the magazine are published exciting pieces written by top leaders in the field as well as students who submit unsolicited work.

I highly recommend reading the issues! Here is the current one.

I am writing this since I have been selected as the Department Editor for this magazine as January 2012. I am looking forward to contribute to this great platform.

I followed a Machine Learning course this semester at the university (along with the one, offered by Stanford Uni). I have been working on a seminar studying Reliable probability estimation in classification using calibration. The final report is in the form of the scientific article, which is attached below.

Here is the abstract:

Estimating reliable class membership probabilities is of vital importance for many applications in data mining in which classification results are combined with other sources of information to produce decisions. Other sources include domain knowledge, outputs of other classifiers or example-dependent misclassification costs. We revisit the problem of classification calibration motivated by the issues of the isotonic regression and binning calibration. These methods can behave badly on small or noisy calibration sets, producing inappropriate intervals or boundary generalization. We propose an improvement of the calibration with isotonic regression and binning method by using bootstrapping technique, named boot-isotonic regression and boot-binning, respectively. Confidence intervals obtained by repeatedly calibrating the set sampled with replacement from the original training set are used for merging unreliable or too narrow calibration intervals. This method has been experimentally evaluated with respect to two calibration measures, several classification methods and several problem domains. The results show that the new method outperforms the basic isotonic regression and binning methods in most configurations.

In November 2011, a show called Ugriznimo znanost (Bite the Science) has been shot at the Faculty of Computer and Information Science by the national public broadcasting organization (RTVSLO).

Bite the Science is a weekly series on the educational channel TV Slovenia. It is the series which aims to explain the science in a relaxed and humorous manner. The topics range from current events in science to accomplishments of Slovene scientists and scientific research in other parts of the world.

The episode scheduled for the last week of December has discussed programming. In it I am presenting my achievements at the Google Summer of Code program (Nimfa library) and discuss Orange, open-source data mining and visualization tool developed by Biolab, Bioinfomatics Laboratory. Further, researchers talk in the episode about smart house, Faculty Robot League, summer school of programming and mathematical data analysis used for predicting the results of the Eurosong competition.

Traditionally Week of The University in Ljubljana is organized in the beginning of December, during which numerous ceremonial events are held at the university and it's faculty members.

This year I was awarded Best Student award of University for overall performance in courses (slo. Svečana listina za izjemen študijski uspeh Univerze Ljubljana) in addition to faculty award which I have received in last years. The reception at the University was well organized with few music intermezzos between speeches.

In this short post I would like to update you on some recent changes in the MF - Matrix Factorization Techniques for Data Mining Library.

Recently, the library MF - Matrix Factorization Techniques for Data Mining has been moved and renamed. It is now called Nimfa - A Python Library for Nonnegative Matrix Factorization or short, nimfa. The latest version with documentation, link to source code and working examples can be found at nimfa.biolab.si. Those of you still having old links, are kindly requested to update the new information - you will be automatically redirected to the new site if you visit old page.

I have always been fascinated by the world of fractals and have been deeply enthusiastic exploring the maths behind them. This post is announcing the support of the fractal dimension computation in the MF - Matrix Factorization for Data Mining library.

In the following paragraphs we shortly revise the most important concepts and definitions of the fractal dimension.

The embedding dimensionality of a data set is the number of attributes in the data set. The intrinsic dimensionality is defined as the actual number of dimensions in which the m-dimensional original vectors can be embedded under the assumption that some distance in the reduced space is kept among them. Given a fractal as a self-similar set of points with self-similar pieces, where each is scaled down by a factor of , the fractal dimension of the object is defined as

.

Example: Sierpinski triangle (My seminar work for CG on L-systems - figure 1 in Appendix of the Report document) consists of three self-similar parts and each is scaled down by a factor of two, therefore its fractal dimension is

For the finite set of points in a vector space, we say the set is statisticallyself-similar on a range of scaleson which the self-similarity is true. However in the theory self-similar object should have infinitely many points because each self-similar part is a scaled-down version of the original object. As a measure of the intrinsic fractal dimension of a data set, the slope of the correlation integral is used. The correlation integral for the data set is defined as

Given a data set which is statistically self-similar in the range, its correlation fractal dimension is

It has been shown that the correlation fractal dimension corresponds to the intrinsic dimension of a data set. Many properties hold for the correlation fractal dimension, see [1] and [2]. For us it is especially important, that the intrinsic dimensionality gives a lower bound on the number on attributes needed to keep the vital characteristics of the data set.

A fast algorithm for the computation of the intrinsic dimension of a data set presented in [2] is implemented in the MF - Matrix Factorization for Data Mining library. Intuitive explanation of the correlation fractal dimension is that is measures how the number of neighbor points increases with the increase of the distance. It therefore measures the spread of the data and the fractal dimension equal to the embedding dimension means that the spread of the points in the data set is maximum.

Of high importance is a Conjecture 1 in [1]: With all the parameters being equal, a dimensionality reduction method which achieves higher fractal dimension in the reduced space is better than the rest for any data mining task. Therefore correlation fractal dimension of a data set can be used:

for determining the optimal number of dimensions in the reduced space,

as a performance comparison tool between dimensionality reduction methods,

and all this can be done in way that is scalable to large data sets.

Recommended reading:

Kumaraswamy, K., (2003). Fractal Dimension for Data Mining. Carnegie Mellon University.

Jr, C. T., Traina, A., Wu, L., Faloutsos, C., (2010). Fast Feature Selection using Fractal Dimension. Science, 1(1), 3-16.

In [1] a concept of intrinsic fractal dimension of a data set is introduced and it is shown how fractal dimension can be used to aid in several data mining tasks. In [2] a fast O(n) algorithm to compute fractal dimension of a data set is presented. On top of that a fast, scalable algorithm to quickly select the most important attributes of the given set of n-dimensional vectors is described.

Google Summer of Code 2011 has finished. On 22th of August it was firm "pencils down" date and today, on 26th of August, has been final evaluation deadline. Therefore, it is time for a small review to be published here on my blog.

I successfully completed the program and have met all the goals, outlined in the original project plan with some (2) additional factorization methods I have implemented. I have been very satisfied with the support and mentoring of both the organization and mentor.

The project, I have worked on, has been developing library MF - Matrix Factorization Techniques for Data Mining which includes a number of published matrix factorization algorithms, initialization methods, quality and performance measures and facilitates the combination of these to produce new strategies. The library contains examples of usage, applications of factorization methods on both synthetic and real world data sets are provided.

Matrix factorization methods have been shown to be a useful decomposition for multivariate data as low dimensional data representations are crucial to numerous applications in statistics, signal processing and machine learning.

An incomplete list of applications of matrix factorization methods includes:

bioinformatics,

environmetrics and chemometrics,

image processing and computer graphics,

text analysis,

miscelllaneous, such as extracting speech features, transcription of polyphonic music passages, object characterization, spectral data analysis, multiway clustering, learning sound dictionaries, etc.

Example using synthetic data set is intended as demonstration of the MF library since all currently implemented factorization algorithms with different initialization methods and specific settings are ran. Others include applications on real world data sets in:

bioinformatics,

text analysis,

image processing,

recommendation systems.

I will outline only the most important content of the MF library here (for any details refer to documentation (or code)), as this is project review and not library reference (references to articles are provided in the documentation).

It has been some time since my last post, but here is the new one. Perhaps the title sounds a bit inappropriate, but indeed it is well suited. Read till the end, where I explain it for those not figuring it yet (or consider it a puzzle :))

So, what have I been up to lately? Despite summer holidays I have been involved in quite a few projects.

First, GSoC Matrix Factorization Techniques for Data Mining project for Orange has been progressing well. Code is almost finished, no major changes in framework, factorization/initialization methods, quality measures, etc. are expected. Project is on schedule and has not diverged from initial plan, all intended techniques (plus a few additional I have found interesting along research) are implemented. I have been doing some testing, and have yet to provide more use cases/examples along with thorough explanation and example data sets. I will not go into details here, as implemented methods' descriptions with paper references are published at Orange wiki project site. The project is great, a mix of linear algebra, optimization methods, statistics and probability, numerical methods (analysis if you want to read some convergence or derivation proofs) with intensive applications in data mining, machine learning, computer vision, bioinformatics etc. and I have been really enjoying working on it, here is my post at Orange blog. The Orange and its GSoCers have been spotlighted at Google Open Source Blog.

Next, there is some image processing; segmentation, primary and secondary object detection, object tracking, morphology measures, filters etc. (no details).

Minor for keeping contact with MS world, Sharepoint Server 2010 (SP 10). I have some experience with it (and its previous version MOSS 2007), both in administration and especially in code. This time it was not about coding workflows using Win Workflow Foundation, developing Web parts/sites/custom content types/web services (...) but providing an in-site publishing hierarchy for data in custom lists and integration with Excel Services (not with new 365 Cloud service). Obstacles were limited server access (hosting plan), old versions of software and usual MS stuffs (:)). In SP 10 these are SPFieldLookups filters and cascading lookups, data connections between sites/lists/other content. As always there are some nice workarounds which have resolved all issues.

Last (not least) I have been catching up with all the reading material I was forced to put aside during the year (well not entirely true: the more I read, the more should be read, so the pile of papers in iBooks and Mendeley app is not getting any smaller :)).

Here we are, what about the post's title? The sunrise problem was introduced by Laplace (french mathematician known for Bayesian interpretation of probability, Laplace transform, Laplace equation, differential operator, work in mechanics and physics). Is the probability that the sun will rise tomorrow 1 if we can infer from the observed data that is has risen every day on record? :) So what is the answer of the question in the title? The inferred probability depends on the record - whether we take the past experience of one person, humanity, or the Earth history. This is the reference class problem - with Bayes any probability is the conditional probability given what a person knows. Simple principle emerged from this, add-one or Laplacian smoothing (Example: Doing spam email classification with a bag of words model or text classification with multinomial model, this allows the assignment of positive probabilities to words which do not occur in the sample) and corresponds to the expected value of the posterior.

Impressions from Google Scholars' Retreat 2011, which took place last month in Zurich, Switzerland, where the Retreat for the EMEA region was organized, are great. It has really been a valuable experience to meet many Googlers, other scholars and their research work.

Below is this year logo.

Let me mention just a few workshops and talks I have attended during the Retreat:

Keynote speech

Working in Industry

Product Design Workshop

Career Panel

Poster Show

Android Technical Talk

Open source and Google Support of its Advancement in the Community (Details: Open source community at Google, development model and how Google uses it and supports advancement of this community.)

Google Web API (Details: Generic principles of Google Web APIs and the tools Google provide for learning and experimenting with them. Focus on interaction with these tools for research, be it to harvest data or to help present data in an efficient way.)

*Privacy, the Ultimate Interdisciplinary Challenge (Details: The questions of deletion and digital rights management, online reputation and self-representation, the right to be forgotten.)

*Google.org - Using Google's Strengths to Address Global Challenges (Details: Google.org is tasked with using Google's strengths in technology and the Internet to address global challenges. The session covered development of Google Flu Trends as well as more recent work including Google Earth Engine and Google Crisis Response.)

*Street View 3D (Details: Many problems arise when developing such a product, such as augmenting the panoramas with 3D features for better navigation and a more immersive experience.)

*Priority Inbox (Details: Because importance is highly personal, email importance is predicted by learning a per-user statistical model, updated as frequently as possible. The challenges include inferring the importance of mail without explicit user labeling; finding learning methods that deal with non-stationary and noisy training data; constructing models that reduce training data requirements; storing and processing terabytes of per-user feature data; predicting in a distributed and fault-tolerant way.)

Beside that there were some very informative talks on business visions of Google, ways of involvement with Google, office tours (I liked the slides between the floors :)) and lots of small talk with Googlers.

One of the courses I attended this semester has been Computer Graphics (CG).

I have spent some time studying algorithmic botany and especially L-systems, formal grammars for describing fractal objects. These can be used for generation of objects in biology, botany, and even buildings and entires cities. Rome Reborn is an example of such project, in which formal grammars were used for the creation of the 3D digital model illustrating the urban development of ancient Rome.

So I have decided to visualize some of the 3D fractal objects using OpenGL and LWJGL library. Below are links to short report and presentation. Take a look :)

Those of you who are interested, great book on this topic by the father of algorithmic botany, Aristid Lindenmayer. Prusinkiewicz, Przemyslaw; Aristid Lindenmayer (1990). The Algorithmic Beauty of Plants (The Virtual Laboratory).Springer-Verlag. ISBN 0-387-97297-8

Yet another great news concering my (little) involvement with Google. I have written few weeks ago about being accepted to the Google Summer of Code 2011 with the project on matrix factorizations techniques in data mining for the Orange platform.

Nevertheless, Google has announcedGoogle Anita Borg Scholarship Recipients and Finalists, a Scholarship for which I have applied this year and I am among 147 undergraduate and graduate students worldwide being chosen. Just for clarification - this is completely unrelated to the GSoC (the only common denominator being the Google itself), the scholarship however being awarded based on the strength of candidates’ academic performance, leadership experience and demonstrated passion for computer science.

Scholars from Europe have Scholars' Retreat at European Google centre at Zurich in June and I am very much looking forward to this event to meet some fascinating people. The retreat will include workshops, speakers, panelists, breakout sessions and social activities scheduled over a couple of days.

(Official Google Blog with published results of the Scholars's selection process) link

(Official Google Students Blog with published announcement of the Scholars's) link

Google Team has shared a new Chrome experiment, called the WebGL Globe, namely the visualization platform for geographic data that runs in WebGL enabled browsers - Chrome, Firefox. (Check http://www.doesmybrowsersupportwebgl.com/ if your browser supports the WebGL standard).

To speed up the visualization of 3D geometry, they have used vertex shader and took advantage of GLSL with two fragment shaders. 3D data spikes are drawn with Three.js, JS library for building lightweight 3D graphics.

I have embedded simple globe showing Google search traffic. Try it or try more examples that shipped with this cool open source project. Or create your own globe using the JSON data format.

Project title: Matrix Factorization Techniques for Data Mining

Description: Matrix factorization is a fundamental building block for many of current data mining approaches and factorization techniques are widely used in applications of data mining. Our objective is to provide the Orange community with a unifed and efficient interface to matrix factorization algorithms and methods. For that purpose we will develop a scripting library which will include a number of published factorization algorithms and initialization methods and will facilitate the combination of these to produce new strategies. Extensive documentation with working examples that will demonstrate real applications, commonly used benchmark data and visualization methods will be provided to help with the interpretation and comprehension of the results.

Main factorization techniques and their variations planned to be included in the library are: Bayesian decomposition (BD) together with linearly constrained and variational BD using Gibbs sampling, probabilistic matrix factorization (PMF), Bayesian factor regression modeling (BFRM), family of nonnegative matrix factorizations (NMF) including sparse NMF, non-smooth NMF, local factorization with Fisher NMF, least-squares NMF. Different multiplicative and update algorithms for NMF will be analyzed which minimize LS error or generalized KL divergence. Further nonnegative matrix approximations (NNMA) with extensions will be implemented. For completeness algorithms such as NCA, ICA and PCA could be added to the library.

Just today I returned from the Portugal, Lisbon, where Euroskills 2010 has been held. In the Office ICT Category (Informatics) I won two silver medals, one as a Project Manager of the team and second together with Slovenian team (members: me, Slavko Zitnik, Miha Longino, Peter Virant).

The contest was organized very well, we were accomodated in the part of the Lisbon, where Expo was held, and there was also the competition. In fifty trades there were 500 competitors and a few hundred of experts, observers and guests.