• Implementation of a system to detect cheaters in programming competitions.
  Programming contests are online events hosting developers from around the world so that they can solve problems in a timely manner. The objective of this theme was to implement a system to calculate the percentage of resemblance of source codes of developers and to produce alerts in case of potential cheats.
• Implementation of an overall score calculation process per user
  CodinGame rankings were defined by competition and programming language in order to identify the best developers. The objective of this theme was to implement a global calculation process in order to achieve a global ranking and an evolutionary graph for each participant.

Adding new programming languages (Haskell, ObjectiveC, VB.Net, Go and Python3) to the CodinGame platform (www.codingame.com).

Content Management Systems (CMS) are websites with a wealth of information for users who want to post or search for informations, however, because of their exponential growth, assigning categories to partially structure information has become uncontrollable . The objective of this work was to evaluate the algorithms and the tools of text analysis and then propose an hybrid algorithm of automatic recommendation of multi-categories and similar contents, from the URL of a website.

Discussion forums are spaces for asynchronous exchanges of text messages. This new mode of communication is very popular with patients because it is associated with a great freedom of speech due in particular to anonymity. The objective of this work was to develop a recommendation system based on an unsupervised classification based on the textual data posted on the forum www.cancerdusein.org.

Crowdsourcing services are increasingly used because they enable low-cost tasks to be carried out by relying on Internet users. The objective of this study was to simulate the evaluation of the results of tasks carried out by users on a Crowdsourcing platform. User profiles were modeled using confusion matrices to represent the knowledge of the user undergoing a classification task, the latter being generated randomly by Monte Carlo simulation.

The linear regression of a variable y of a matrix X of regressors is impossible to do directly in large dimension when the number of regressors in X is greater than the number of observation, in other words, the matrix X′X isn’t invertible, then the regression needs to be regularized. The work consisted in understanding a proposed new approach to regularization based on the extraction of components by penalization, then programming several variants, then testing and comparing them with conventional regression methods (e.g. Ridge, PLS, etc.) on real data.

Methods of economic evaluation in health are a necessary instrument to compare the medical strategies that will be evaluated by health decision maker of the country (HAS in France) so that they can take a relevant decision (i.e. price, reimbursement, and so on). The objective of the internship was separated in two stages; (1) evaluate and analyze the Heemod package by modeling different real clinical decision-making cases in order to certify the choice of the package, (2) validate the results obtained with Heemod against the results obtained in Excel sheets (i.e. official documents per health authorities) in order to ensure the transparency of these results and to promote a new certified tool for the use of health authorities.