First, we generalize their idea for binary classification to multiple classes. Statistical techniques from the field of survival analysis are widely used for the assessment of treatment effects in randomized clinical trials. In this thesis, we aim to find methods for predicting peaks in twitter time-series which correspond to almost simultaneous real- world events. One main focus of the thesis lies on describing different approaches to apply SDP relaxations to solve the non-convex optimization problem. In this thesis, we consider some relaxations of the hidden variable conditions in Frot et al Our experiments show that the developed classifiers outperform the baseline classifiers on all four data-sets. This thesis provides a thorough introduction to topic modeling focusing on latent Dirichlet allocation LDA.

Maathuis Pegah Kassraian Fard Prof. They consist of a box with millions of particles that in- teract with each other due to gravitational forces along cosmic time. Reduction of the absolute additive bias due to dynamical scaling was evaluated by comparing the bias components associated to the RCM-GCM chains and their corresponding drivers. The second estimates the ocurrence of a deforestation event by comparing the last few predicted and real reflectance values. Jun Wu Learning directed acyclic graph with hidden variables Prof. Aim of the thesis is to investigate these approaches and, based on them, analyse intraday covariance dynamics and develop an intraday portfolio risk methodology. In contrast to typical simulation designs, we do not arbitrarily generate new data, but rely on existing datasets on which several different models are fitted.

Genome wide association studies GWAS are exploratory approaches to htesis previously unknown associations between common types of variation in the genome and a phenotype like the presence of a disease or the characteristic of a trait. Prediction error can be bounded in both thesiz. We show that a model trained in a specific swissquxnt can be successfully used for navigation in other unseen environments. Firstly, the mathe-matical description of Critical Line Algorithm is introduced.

In this thesis we discuss two topics. Furthermore, we analyze whether there have been any changes in the reporting of data breach events due to the introduction of data breach notification laws in the US. Identifying financial bubbles and predicting the burst of them is of high theoretical but also practical interest. The goal of this thesis is to model real estate prices with deep learning. Optimal mass transportation is on its way to become a major tool in numerous fields of application – in particular in machine learning.


Look maaster further and apply now! Classic prediction relies on the assumption that the data generating mechanism does not change in between training and prediction.

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. We compare two methods which test hierarchically groups of predictor variables for signif- icant association with a response variable swlssquant a high-dimensional setting while controlling the family-wise error rate.

swissQuant Group AG

Also, the increase in power does not come at an expense of higher error rates — we were able to achieve similar error rates when aggregating p-values or when pooling the data. Moreover, we provide complementaries regarding the relevant literature of covariance matrix estimation and changepoint detection in similar settings, thfsis parameter selection, models for simulations and error measures to evaluate performances.

master thesis at swissquant

This database includes resting state functional and structural MRI scans. 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.

In this project, we implement a model for mortality prediction of pa- tients hospitalized in the Intensive Care Unit based on their clinical notes. In comparison to Kassraian Fard et al. This is surprising as the models learn the meaning of words solely by being trained on plain text data.

This allows us to gain deeper insights in the general behaviour of such networks. We conclude by discussing some of the more recent developments in swiasquant field and provide a small application of the results studied in Appendix B.


This is compared to the performance of the models trained on an unbalanced swissquaht set. Deep reinforcement learning has also been applied in the field of robotics, enabling robots to learn complex behavior directly from raw sensor inputs. The modification of k-means using Shape-Based distance and a related centroid function returns the best partitions, based on internal evaluation indexes, namely Average Silhouette Width and COP Index. We consider the problem of finding the structural breaks, also referred to as changepoints, in a sequence of non-homogeneous, high-dimensional data.

While population average models rely on generalized estimating equations, generalized linear mixed models use restricted maximum like-lihood for computing parameter esti-mates. We present a rigorous comparison with well-known information retrieval approaches such as bag- of-words BOWtf-idf and swiwsquant dirichlet allocation LDA.

However, due to some problems regarding data-quality and since some of the methods we used are only suitable for retrospective views, our results should be followed up on with caution. Given the difficulty of addressing the problem at hand from a theoretical perspective, we approach it empirically. This thesis presents the theory and main ideas behind some of the nowadays most popular methods used for causal structure learning as well as the ICP algorithm, a new algorithm based on a method recently developed at ETH Zurich.

We will call this approach the Jacobian dimension estimation. Therefore, the predictive power is limited.

Emplois : Swissquant, Zürich, ZH – mai |

In this thesis we are analysing three sets of financial time series with R. Our results furthermore indicate, that the advantage masger learned image compression with perceptual loss functions is especially pronounced below 0.

master thesis at swissquant

For the application to the nearest neighbor search problem we will see that the Johnson-Lindenstrauss Lemma represents the bottleneck in terms of time re-quired.