We analyse the structure of the different models in detail, show relationships between them and further simple models and proof how additive compositionality arises in certain models under some constraints. Throughout the thesis, additional results are derived to facilitate the understanding of the in- volved theorems and definitions. This is made precise by the notion of em- pirical measure, which is a proper probability measure for each realization. The advantages of our algorithm are more pronounced in a dense setting. This problem inspires to think about comparing risk of estimators given data from two different distributions. In Markowitz’s Modern Portfolio Theory, portfolio on Efficient Frontier has the minimized risk at a given level of expected return, or has the maximized expected return at a given level of risk.

Our model achieves a Mean Reciprocal Rank of 0. In Zhu et al. Such networks can be modeled by using Holarchy, a hierarchical self-organization technique using autonomous agents who also serve as the part of the network. These methods are particularly intended for high-dimensional data sets where the dimension of the variables is comparable to or even greater than the number of available training data samples. It is also shown that among all asymptotically optimal valid adjustment sets, O yields the strictly best finite sample variance. Based on the idea of the paper from Heuvelink et al.

Our mastet achieved extraordinary perfor-mance in the validation set across all acronyms and remarkable generalization power in an independent dataset. Moreover, its performance does not depend on additional tuning parameters. Supporting the establishment of global marketing plan and its execution Run analysis in SAP and Excel e. In this thesis, we aim to propose a novel multilayer non-linear approach to a variant of NMF, that is able to learn such hidden attributes.

In this thesis, we propose a totally automated and unsupervised mitral valve segmentation algorithm. Our business edge originates from the effective translation of Intelligent Technology into measurable, bottom-line client value. In the second part it is shown how these challenges can be tackled.

Emplois : Swissquant, Fahrweid (nördl. Teil) / Fahrweid, ZH – mai |

The pre-optimization and simultaneous execution of these models provide a higher degree of freedom in the system in search of the optimum. Furthermore, the classification procedures illustrated awissquant this thesis could be used to repeat the swissquajt study and reach thesls comparable results.


We propose many experimental procedures and implement simulations to test their ability in controlling the FDR. Polina Minkina A new hybrid approach to learning Bayesian networks from observational data Dr.

An extension to binary segmentation called optimistic binary segmentation is proposed that, to the best of our knowledge, is the first approach with a logarithmic, and hence sub-linear number of required model fits. Marloes Henriette Maathuis Jul Abstract: However, on the KOF dataset, a generalised linear mixed-effect model could outperform the baseline and was able to classify shrinking and non-shrinking restaurants.

Topic The thesis will focus on the estimation of high-frequency covariance dynamics for financial assets, with the purpose of developing a methodology to assess intraday portfolio risk figures.

We show that our system performs competitively with other systems on thesiis standardized separation task. A scalable approach to changepoint detection in high-dimensional graphical models Prof. While population average models rely on generalized estimating equations, generalized linear mixed models use restricted maximum like-lihood for computing parameter esti-mates.

Two variants of this algorithm and eight additional statistical or machine learning methods are compared against each other, in terms of six evaluation metrics on their out-of-sample performance. After studying all of these, we try to make use of the nodewise and fixed-X knockoff ideas, and find sswissquant method that guarantees the finite sample FDR control when learning the structure of a Gaussian graphical model.

master thesis at swissquant

To illustrate that this is not the case for the Cox hazard ratio in a setting with unmodelled heterogeneity, we reproduce theoretical results as well as simulation studies from Aalen et al.

Motivated by previous work, in particular the work of Eling and Loperfido [14, ], Wheatley, Maillart and Sornette [37, ], Hofmann, Wheatley and Sornette [18, ], we analyze data breaches with at least 70k records lost from an insurance point of view with a new extended dataset.

In the first prediction errors form first model for selected bands are first compared against, band-specific, thresholds to produce one deforestation flag per band.


swissQuant Master Thesis Job in Zürich | Glassdoor

We also include a simulation section in this thesis, where we do simulations on CTE models: The fitted object is used to generate new response variables.

Traditional approaches to solving this problem were typically non-data-driven, using known statistical properties of music signals to perform the decomposition. For that, we assume that for a topic of interest the twitter-network in some way undergoes a bifurcation whenever or shortly before the corresponding time-series of tweet-counts reaches a peak. Digital parking space management and the Internet of Things IoT enable data collection of occupancy rates in parking lots through sensor technology.

The computation of these interactions in an accurate way is expensive Teyssier et al. The first step removes the hidden variables with an effect on a large proportion of the observed variables.

The main result that is achieved by this work is related to the positive correlation between Life Premium and Gross Domestic Product: This thesis aims to improve the statistical estimation of the LPPLS model by allowing the residuals of the parametric model to have an auto-regressive part and heteroskedasticity in the inno- vations.

swissQuant Group AG

We present a rigorous comparison with well-known information retrieval approaches such as bag- of-words BOWtf-idf and latent dirichlet allocation Tgesis. Thus, the optimization problem is relaxed to the semi-definite programming SDP framework and its solution can then be approximated or sometimes even obtained exactly in polynomial time.

The dimensionality and presence of many highly correlated predictors make building truly accurate models and their interpretation particularly chal-lenging.

master thesis at swissquant

We run simulation studies and compare performance of our approach to other algorithms for structure learning, such as PC-algorithm, greedy equivalent search GES and max-min hill climbing MMHC.

Emilien Jules Generalized Linear Models: