Hey there! My name is Gilian Ponte. I am an Assistant Professor at Rotterdam School of Management, Erasmus University Rotterdam. In my research I try to address societally relevant issues in marketing. Currently, I work on the balance between privacy protection (using differential privacy) and the ability to derive insights from data. For example, is it still possible to derive profits from targeting under a privacy protection guarantee for consumers? For the near future, I would like to study misinformation and online echo chambers.
I obtained my PhD at the University of Groningen. My two excellent promotors during my PhD were Jaap Wieringa and Tom Boot.
Previously, I have worked for two years as a web analyst at Blokker Holding (a major retailer in the Netherlands).
Probably about 70%-90% we write as academics does not end up in journals. In this blog, I would like to share the things that I think are still interesting to share with a more general audience. Some pieces are more research heavy while in other posts I will try to address practitioners or students. Please be aware that some of the posts below were written a very long time ago. I keep them on here to observe the progress over time.
Posts
A very simple membership inference attack.
20 March, 2022
How GANs learn probability density functions.
7 September, 2021
Generative adversarial networks (GANs): generating celebrity faces.
27 March, 2020
As part of the Deep Learning course at Rijksuniversiteit Groningen, we aim to generate celebrity faces from the CelebA data set. We experienced that for images over 64x64 pixels serious computing power is required. Also, we experimented with different architectures. An overview is available at my YouTube channel or for code my Github.
Predicting customer churn: Which estimation method should I use?
26 June, 2018
Author(s): Gilian Ponte
This article presents a comparative study of churn estimation methods. Telecommunication providers can no longer rely on a steady customer base. Machine learning methods are applied to the problem of customer churn in the telecommunications industry. In the first section, relevant variables for explaining churn behaviour are evaluated. Followed by a description of the methodology of a logistic regression, decision trees, bagging, boosting and a neural network to model churn behaviour. These models are estimated and evaluated by comparative performance measures. The results show that bagging performs better than decision trees, boosting, neural network and a logistic regression. Will a neural network outperform the more classical approaches?
Scraping with R
16 October, 2017
Author(s): Gilian Ponte
Scraping is a time-saving skill. It makes you save time for more important things in your daily routine work or hobbies (for example: getting coffee for your colleagues). More important in my experience, for most companies it enables the company to analyse its competitors’ pricing strategy, product availability or collect reviews to do a sentiment analysis. In this case, we will scrape some prices from a Dutch webshop (please don’t sue me).