pierre haenecour phd thesis

to one time step. Hassouni Afrae, "Assessing Portfolio Performance Using a Non-parametric Approach Co-directors: Laurent Gheeraert and Hugues Pirotte. Tsatsis Christos, "Three Essays in Public Accountancy Director: Faska Khrouz. Thus, inspired by the high variance of tree parameters, we propose a new algorithm based on extremely randomized trees (extra-trees).

This method is very successful in terms of accuracy but does not provide fully interpretable models. With 4000 doctoral students out of a total of 18'500 students, ETH Zurich is one of the universities in Europe which focuses most intensively on research. We start our investigations by an empirical study which shows quantitatively how important decision tree variance is,.e. In order to also recover interpretability, we then propose a dual' perturb and combine algorithm which randomizes input attributes at the prediction stage while using one single model. These experiments confirm that the rules induced for a given problem are indeed highly variable from one learning sample to another, and thus that variance is detrimental not only for accuracy but also from the point of view of interpretability. Apart from a few exceptional cases, all doctoral students at ETH Zurich are employed as scientific assistants. Decision tree induction is a supervised learning algorithm, which focuses on the modeling of input/output relationships in the form of if-then rules. Marie, gorce Reviewer Pr of essor, insa-Lyon - France Thierry Turletti Reviewer Researcher, inria - Sophia Antipolis - France Hakima Chaouchi Examiner Pr of essor, Institut Télécom SudParis - France. De Brouwer Octave, "The Impact of Globalization and Economic Cycles on Worker's Health and Disability Benefits Director: Ilan Tojerow.

pierre haenecour phd thesis

Graduate school penn state thesis
Minor corrections phd thesis
Pequod printing thesis