Master Thesis: A Study of Urban Complaints in Twitter. Grade 6/6.Collected 311 complaints (reference dataset) and tweets from the New York area. Developed a method to expedite the labelling task based on bootstrap labels and Logistic Regression classifier. Trained a 7-classes SVM-RBF classifier to detect 311 complaints in tweets, achieving < 1% FPR despite the heavy class imbalance. Conducted a qualitative comparison of the two complaint channels based on the frequency, temporality, and spatiality

Topic: Predicting the Popularity of YouTube Videos.Experimented with Affinity Propagation and Gaussian Mixture models for piecewise time-series clustering, and Bayesian Network for predicting video views.

Taught Java as part of a humanitarian project led by Prof. RD Hersch. Wrote the course material, prepared the hands-on exercises, conducted classes, and assisted students with their coding exercises.

Performed data modelling, management, and data mining tasks for a research project commissioned by an external company

Development and testing on GAPS, an academic management tool for students and professors

Developed a specialized GIS software for creating thematic maps for the Federal Statistical Office. The work was presented at FOSS4G 2007 in Victoria, Canada

Implemented a Neural Network trained with a Genetic Algorithm to detect deforestation from satellite images and climate data. This work was presented at the 6th European Conference on Ecological Modeling in Trieste, Italy and received honors from the thesis committee