# [MVA Master](http://www.math.ens-cachan.fr/version-francaise/formations/master-mva/) - 2015-2016 ([Lilian Besson](https://perso.crans.org/besson/))
Documents produced for my [2nd M.Sc.](https://perso.crans.org/besson/cv.en.pdf) in *applied mathematics*, *machine learning* and *optimization* (in 2015-16).
## Rank and grades
- The work I did during my internship was evaluated to **19 / 20**. But what's important is the three following points, roughly: a) swimming every day in Lausanne's lake was awesome, b) the main theorem proved during my internship boils down to [the Fundamental Theorem of Algebra for polynomials with real-valued coefficients](https://en.wikipedia.org/wiki/Fundamental_theorem_of_algebra#Corollaries), c) [my master thesis](https://goo.gl/xPzw4A) starts with a fun and pretty a 30-pages introduction that I loved writting (the rest was more boring).
- For the courses, my average grade is **18.14 / 20** (for the *best 8 grades*, among the 13 courses I validated, giving the grades *19*, *19*, *18*, *18*, *18*, *18*, *17.64*, *17.5*).
> Apparently, my record remains unchallenged, and this has absolutely no interest: you (and I and everybody) should study **to learn**, not for grades!
- ⇒ so my final grade for the MVA M.Sc. is **18.43 / 20**. Sweet.
- I got [*summa cum laude*](https://en.wikipedia.org/wiki/Latin_honors) ("très bien avec félicitations du jury").
- I also **ranked first**, both in term of number of validated courses (**13** courses, instead of the minimum 8), and in term of the [average grade](https://en.wikipedia.org/wiki/Academic_grading_in_France) (**18.43 / 20**).
## Other documents, and documents shared here.
> - See also [some documents from GitHub](GitHub/), or [my resume](https://perso.crans.org/besson/cv.en.pdf) or [my Bitbucket profile](https://bitbucket.org/lbesson) if needed.
> - For both semester and for each course, I added my grade, the average grade and the number of students who validated the course (note: the standard deviations are between 2 and 3 points /20 for each course).
> - Only some PDF documents are shared here, the code for homeworks and TP will *not* be shared, but the code for all my small research projects and my internship *is* available (and open-source).
## **Research internship** (Spring and Summer 2016)
> Grade: **20/20** given by my advisors (best possible grade), and **19/20** given by the MVA jury.
- I was a visiting research intern at EPFL, [for a 4.5-month research internship](Sujet_de_stage__MVA_16__Lilian_Besson__LIB_EPFL_Lausanne.en.pdf) in the [BIG team (STI, EPFL)](http://bigwww.epfl.ch/people.html), from 15th of April to 30th of August 2016, to conclude my M.Sc.
- I worked on operators theory and functional analysis, mainly on convolution operators and their properties (homogeneity, rotation-invariance, steerability, unity).
- The [slides for my oral presentation](https://goo.gl/vm8WPF) are here: https://goo.gl/vm8WPF (40 min),
- And [my master thesis / internship report](https://goo.gl/xPzw4A) is here: https://goo.gl/xPzw4A (~ 100 pages),
- All these documents (and their LaTeX sources) [are fully open-source](https://bitbucket.org/lbesson/internship-mva-2016/) !
<a href="#list-of-courses"><em id="list-of-courses">¶</em></a>
## List of courses
## **1st trimester** (Fall 2015)
### [Graphs in Machine Learning (GML)](http://researchers.lille.inria.fr/~valko/hp/mva-ml-graphs.php) course
> **18/20**, average *15.8/20* for *47* students.
- My reports for [*TP #1*](MVA_2015-16__Graphs_in_Machine_Learning__TP1__Lilian_Besson.en.pdf), [*TP #2*](MVA_2015-16__Graphs_in_Machine_Learning__TP2__Lilian_Besson.en.pdf) and [*TP #3*](MVA_2015-16__Graphs_in_Machine_Learning__TP3__Lilian_Besson.en.pdf) (no code).
- For the [GML project on "Inverse Reinforcement Learning for Board Game Inference"](https://bitbucket.org/lbesson/mva15-project-graph-reinforcement-learning): [our slides](MVA_2015-16__GML_and_RL__Project__Lilian_Besson__Basile_Clement__Slides_19-01-16.en.pdf), [our report](MVA_2015-16__GML_and_RL__Project__Lilian_Besson__Basile_Clement__Final_report.en.pdf) (I worked with [Basile Clement](http://www.eleves.ens.fr/home/bclement/CV.pdf) from [ENS Ulm](http://www.ens.fr/)). Our advisors were [Émilie Kaufmann](http://www.di.ens.fr/~kaufmann/) and [Christos Dimitrakakis](http://www.cse.chalmers.se/~chrdimi/); January 2016.
### [Convex Optimization (CvxOpt)](http://www.di.ens.fr/~aspremon/OptConvexeM2.html) course
> **18/20**, average *14.10/20* for *87* students (yep, it's a popular course!).
- My report for [*Homework #3*](MVA_2015-16__Convex_Optimization_HM3__Lilian_Besson.en.pdf), on *« Sub-Gradient, Coordinate Descent and Proximal methods for LASSO and Non-Convex Optimization »* (no code); November 2015.
- *Final Exam* summary sheet (very detailed): [one-column (3 pages)](MVA_2015-16__Convex_Optimization_Final_Exam_help_notes__Lilian_Besson.en.pdf), [two-column (2 pages)](MVA_2015-16__Convex_Optimization_Final_Exam_help_notes__Lilian_Besson.2col.en.pdf).
### [Sparsity and Compressed Sensing (PCS)](http://gpeyre.github.io/teaching/) course
> Grade: **19/20**, average *14.19/20* for *37* students. **I ranked first for this course!**
- My reports for [*TP #2*](MVA_2015-16__Compressed_Sensing__TP2__Lilian_Besson.fr.pdf), [*TP #3*](MVA_2015-16__Compressed_Sensing__TP3__Lilian_Besson.fr.pdf), [*TP #4*](MVA_2015-16__Compressed_Sensing__TP4__Lilian_Besson.fr.pdf) and [*TP #5*](MVA_2015-16__Compressed_Sensing__TP5__Lilian_Besson.fr.pdf) (no code).
- For the [PCS project on "Probabilistic Algorithms for Approximate Matrix Decompositions"](https://bitbucket.org/lbesson/mva15-project-parcimonie-compressed-sensing/): [my slides](MVA_2015-16__Compressed_Sensing__Project__Lilian_Besson__Slides.en.pdf), [my report](MVA_2015-16__Compressed_Sensing__Project__Lilian_Besson__Report.en.pdf); January 2016.
- I also presented this project, with [these slides](Lausanne__08-02-16__Lilian_Besson__Internship_2016.en.pdf) at an internal [lab meeting](http://bigwww.epfl.ch/people.html?photo=2015) at the [BIG (LIB) team](http://bigwww.epfl.ch/) at [EPLF, Lausanne, Switzerland](http://www.epfl.ch/) the 08-02, and [some of them](http://bigwww.epfl.ch/storath/index.html) seemed to like it.
### [Reinforcement Learning (RL)](http://researchers.lille.inria.fr/~lazaric/Webpage/MVA-RL_Course15.html) course
> **18/20**, average *14.9/20* for *72* students (a popular course too!).
- My PDF reports for [*TP #1*](MVA_2015-16__Reinforcement_Learning__TP1__Lilian_Besson.en.pdf), [*TP #2*](MVA_2015-16__Reinforcement_Learning__TP2__Lilian_Besson.en.pdf), [*TP #3*](MVA_2015-16__Reinforcement_Learning__TP3__Lilian_Besson.en.pdf) and [*TP #4*](MVA_2015-16__Reinforcement_Learning__TP4__Lilian_Besson.en.pdf) (no code).
- For the [RL project on "Inverse Reinforcement Learning for Board Game Inference"](https://bitbucket.org/lbesson/mva15-project-graph-reinforcement-learning): [our slides](MVA_2015-16__GML_and_RL__Project__Lilian_Besson__Basile_Clement__Slides_19-01-16.en.pdf), [our report](MVA_2015-16__GML_and_RL__Project__Lilian_Besson__Basile_Clement__Final_report.en.pdf) (I worked with [Basile Clement](http://www.eleves.ens.fr/home/bclement/CV.pdf) from [ENS Ulm](http://www.ens.fr/)). Our advisors were [Émilie Kaufmann](http://www.di.ens.fr/~kaufmann/) and [Christos Dimitrakakis](http://www.cse.chalmers.se/~chrdimi/); January 2016.
### [Probabilistic Graphical Models (PGM)](http://www.di.ens.fr/~slacoste/teaching/MVA_GM/fall2015/) course
> **15.5/20**, average *14.1/20* for *91* students (popular course!), 101 were registered.
- My PDF report for [*Homework 1*](MVA_2015-16__Probabilistic_Graphical_Models__DM1__Lilian_Besson.en.pdf), on *« Learning in Discrete Graphical Models »*, and on *« Linear Classification »* -- LR, LDA, QDA methods (no code).
- PGM *Final Exam* help sheet (short): [one-colum (one page)](MVA_2015-16__Probabilistic_Graphical_Models__Final_Exam_help_notes__Lilian_Besson.en.pdf).
- For the [PGM project on "Hidden semi-Markov Models"](https://bitbucket.org/lbesson/mva15-project-probabilistic-graphical-models): [our poster](MVA_2015-16__PGM__Project_on_HSMM__Besson_Brunck__Poster_06-01-16.pdf), [our report](MVA_2015-16__PGM__Project_on_HSMM__Besson_Brunck__Report_13-01-16.pdf) and [the code](MVA_2015-16__PGM__Project_on_HSMM__Besson_Brunck__Code_Figures__13-01-16.zip) (I worked with [Valentin Brunck](http://www.phytem.ens-cachan.fr/annuaire/brunck-valentin-158738.kjsp) from [ENS Cachan](http://www.ens-cachan.fr/)); December 2015.
- I went to [MLSS 2016](http://learning.mpi-sws.org/mlss2016/) to present [this poster](MLSS_2016__Poster_on_HMM_and_HSSM__Lilian_Besson__05-2016.en.pdf), in Cadiz (Spain) in May 2016!
### [Introduction to Statistical Learning (ISL)](http://nvayatis.perso.math.cnrs.fr/ISLcourse-2015.html) course
> **18/20**, average *13.6/20* for 59 students.
- No document were produced for this *purely theoretical* course. But it was a very interesting course!
## **2nd trimester** (Spring 2016)
### [Functional Cerebral Imaging and Brain-Human Interface (fMRI)](http://www.math.ens-cachan.fr/version-francaise/formations/master-mva/contenus-/imagerie-fonctionnelle-cerebrale-et-interface-cerveau-machine-161979.kjsp)
> **17.64/20**, average *11.01/20* for 14 students.
- Practical Session [« Convergence of the Space-Net classifier »](https://bitbucket.org/lbesson/mva16-tp-brain-imaging-nilearn) to decode brain images from the Haxby 2001 experiment: [my PDF report](MVA_2015-16__Brain_Imaging__TP__Lilian_Besson.en.pdf) and [my code](https://bitbucket.org/lbesson/mva16-tp-brain-imaging-nilearn/src/master/TD11__Convergence_of_the_Spacenet_classifier.py); March 2016.
### [Kernel Methods for Machine Learning (Kernel)](http://lear.inrialpes.fr/people/mairal/teaching/2015-2016/MVA/)
> **17.3/20**, average *14.7/20* for 148 students (*very popular* course, as it is available for others Master programs).
- Homework: [PDF report for the Kernel Methods homework](MVA_2015-16__Kernel_Methods__Homework__Besson_Clement_Zerbib.en.pdf), and [this small script](https://bitbucket.org/snippets/lbesson/ay5yE) designed to find numerical counter-examples for non-positive-definite kernel functions K(x,y); February 2016 (I worked with [Basile Clement](http://www.eleves.ens.fr/home/bclement/CV.pdf) and [Nissim Zerbib]() from [ENS Ulm](http://www.ens.fr/)). We got *20/20* (best grade).
- [Kaggle Data Challenge : « Recognizing hand-written digits with SVM »](https://bitbucket.org/lbesson/mva16-kaggle-data-challenge-kernel-methods): [our code](https://bitbucket.org/lbesson/mva16-kaggle-data-challenge-kernel-methods/src/master/src/) and [our PDF report](MVA_2015-16__Kaggle_Kernel_Methods__J_Roquiro__L_Besson__M_Ralle__22-03-16__Rapport.fr.pdf); March 2016 (I worked with [Jaime Roquiro]() from [ENS Ulm](http://www.ens.fr/) and [Mathilde Ralle]() from [Université Paris-Sud](http://www.u-psud.fr/)). We got *14.6/20*.
### [Statistics in High Dimension (SHD)](http://certis.enpc.fr/~dalalyan/MVA2016.html)
> **17/20**, average *12.79/20* for 48 students.
- 2 homeworks: hand-written and scanned, useless to upload them here. But it was a very interesting course!
- Final exam: *nothing here*.
### [Prediction for Individual Sequences (SeqL)](https://sites.google.com/site/vianneyperchet/mva) (*"Sequential Learning with Bandits algorithms"*)
> **19/20**, average *15.58/20* for 12 students.
- A few comrades and I were in charge of writing the lecture notes (8 lectures), now available on [the course website](https://sites.google.com/site/vianneyperchet/mva).
- Final exam: *nothing here*.
### [Modelisation in Neuro-Sciences (MNS)](http://www.lps.ens.fr/%7Enadal/Cours/MVA/)
> **17.5/20**, average *15.9/20* for 13 students.
- For the neuro-sciences project: [*« Self-Organizing Maps (SOM) and Dynamic SOM, From unsupervised clustering to models of cortical plasticity »*](https://bitbucket.org/lbesson/mva16-project-modelisation-neuro-sciences/) : [my report](MVA_2015-16__Neuro-Sciences__Project__Lilian_Besson__Report.en.pdf) and [the slides](MVA_2015-16__Neuro-Sciences__Project__Lilian_Besson__Slides.en.pdf); based on [Nicolas P. Rougier](http://www.labri.fr/perso/nrougier/)'s work; March 2016.
### [Stochastic Images - Signal Analysis (StoImag)](http://www.math.ens-cachan.fr/version-francaise/formations/master-mva/contenus-/methodes-stochastiques-pour-l-analyse-d-images-162028.kjsp)
> **13/20**, average *14.4/20* for 29 students (I did *not* follow the course seriously, I just took the final exam and did the homework very quickly!).
- 1 homework: hand-written and scanned, useless to upload it here.
- Final exam: *nothing here*.
### [Simulation-based learning: theory and algorithms (MCMC)](https://mvamcmc.wordpress.com/)
> **11/20**, average *11.20/20* for 49 students (I did *not* follow the course seriously, I just took the final exam!). But it was a nice course!
- Final exam: *nothing here*.
All these documents are my property, © 2015-16, [Lilian Besson](https://perso.crans.org/besson/) (if not explicitly stated otherwise).
These documents are publicly released under the conditions of [the MIT open-source license](http://lbesson.mit-license.org/).
<noscript><img alt="GA|Analytics" style="visibility:hidden;display:none;" src="https://ga-beacon.appspot.com/UA-38514290-1/publis/mva-2016/HEADER.md?pixel"/></noscript>
> List of documents:
# [MVA Master](http://www.math.ens-cachan.fr/version-francaise/formations/master-mva/) - 2015-2016 ([Lilian Besson](http://perso.crans.org/besson/publis/mva-2016/))