We use multidi- mensional scaling to identify severity risk classes based on the economic sector. The following classification methods were compared in a simulation study: This study investigates the impact of online reviews in the gastronomy industry and its predictive capacity on restaurants financial performance as measured by revenue and growth. Statistical techniques from the field of survival analysis are widely used for the assessment of treatment effects in randomized clinical trials. 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. The qualitative analysis of the 39th and the 44th legislative period unveiled important socio-economical and political events of these time periods. Today, paper has been replaced by PDF documents in many areas because it provides a cheap mean to archive large amounts of documents.

The first step is the identification of events as peaks on time series describing the numbers of tweets published with specified hashtags, hence this problem falls in the area of anomaly detection. There have been several attempts to prioritize relevant tissues and target genes by computing colocalization posterior between GWAS and expression Quantitative Trait Loci eQTL , but biological knowledge has not been fully considered to model the causal status in these cases. To construct the models for this counter-factual analysis, several machine learning procedures are used and compared, selecting the ones that perform the best with respect to out-of-sample predictions. The goal is to find a dictionary from a trainingset of facial images that allows us to represent these images as a sparse linear combinationof the dictionary elements. In this case, LDA and logistic regression were found to have the best cross-validated performance. This approach allowed us to make conclusions about how the methodology generalizes accross space specifically pixels and accross the day of the year. In general, when dynamical systems come close to a tipping- point, time-series that originate from it are expected to exhibit certain early-warning- signals EWS.

Emplois : Swissquant – mai |

This constitutes a major improvement over the baselines, which do not provide interpretable results and solely rely on word semantics. Though these approaches are extensively investigated in research community, it is still limited to different variants of matrix factorization. The following R packages were used: We estimate models with pyramid and at layer composition as well as networks with embeddings for macro location.


master thesis at swissquant

Following an introduction to the univariate ROC curve, estimators mzster the one-dimensional case were compared under a variety of simulation settings. The following qualifications are prerequisites: When testing for significance in high-dimensional datasets such as when dealing with genome-wide datasetsmultiple-testing becomes an inherent problem. Finally, the classification methods were applied to a medical data set.

A Machine Learning Approach Prof. These univari- ate estimators are: This is mastef by testing their classification performances beyond the framework in which the cognitive data were recorded, showing consistent improvement in the detection of sentiment on the Stanford Sentiment Treebank corpus with multiple variants of said cognitive embeddings.

The findings are then applied to a dataset of the Swiss National Forest Inventory http: Francesco Ortelli Statistics meets optimization: To this end, evidence of the potential of EEG and ET, as enhancers of Machine Learning systems, is provided by leveraging such data to improve the detection of sentiment with respect to a baseline model.

Critical Line Algorithm is of great importance due to the fast implemen-tation compared with other optimization approaches. 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.

Navigation Area

We thesjs that a model trained in a specific environment can be successfully used for navigation in other unseen environments. Reinforcement Learning, aided by the representation learning power of deep neural networks has enabled researchers to solve complex decision making problems, the most notable one being AlphaGO, a computer program which beat the champion of the board game GO.

Its competitive prediction performance underpins the promising direction of aggregating randomized low-dimensional projections.

master thesis at swissquant

This thesis aims to develop a methodological framework for predicting parking lot occupancy rates for the city of Zurich empirically. In case of a linear data generating mechanism, anchor regression is recalled and illustrated with examples. This index will measure how atypical a given multivariate predicted mastfr is therby helping us to detect when an intervention to the data generating mechanism has occurred, i.


This includes both partitional and hierarchical techniques with specific distance measures for time series – such as Shape-Based distance, Dynamic Time Warping and a distance derived from the TOPS Symmetric Thermal Optimal Path method. Nino Antulov-Fantulin Kaster Abstract: We high-light the important properties of the methods and develop an improved version of Hierarchical Inference approach.

Lukas Hofmann Estimating the causal e ffect of case management on healthcare expenditure Prof. We close this thesis with a summary and some possible future research directions. In contrast to conventional methods like lasso or ridge regression, this method is able to discover signals, which are neither sparse nor dense. In general, when dynamical systems come close to a tipping- point, time-series that originate from it mastef expected to exhibit certain early-warning- signals EWS.

In between the swissquany the data are assumed to be identically distributed, e.

In this thesis, we mainly discussed a typical problem in express companies’ operations management. In recent years, data-driven approaches, and deep neural networks in particular, have delivered a radical improvement over traditional approaches towards this goal.

The main result that is achieved by this work is related to the positive correlation between Life Premium and Gross Domestic Product: Long Short-Term Memory and other gated networks combined with gradient clipping strategies have been successful at addressing these issues.

The last problem is to get rid of prefixes and suffixes in the classified words, so that only the information contained in the word is kept. Finally, we test the algorithms in the presence of hidden confounders.

master thesis at swissquant