PhD Position in Causal Data Science
Jheronimus Academy of Data Science (JADS), Den Bosch (4 year project)JADS, a data science institute founded jointly by the University of Tilburg and the Technical University of Eindhoven, is looking for a talented PhD student working on 'Causal Data Science', starting as soon as possible. The project is funded by JADS internally.
The PhD-candidate will be employed at Tilburg University (TiU) at the department of the professor who will act as prime PhD supervisor/promotor.
The Jheronimus Academy of Data Science (JADS)
is the unique collaboration of two renowned universities bringing together their top academics to further the emerging and multidisciplinary academic field of data science. JADS not only offers nine academic educational programs (Bachelor, Master, PDeng, PhD and Professional) but also hosts three research centers; the Data Science Center Eindhoven (DSC/e), the Data Science Center Tilburg (DSC/t), and the Data Science Center Den Bosch (DSC/db) at the new Mariënburg campus. At these centers we conduct groundbreaking research on data science and create concrete societal value. Project description
Data science has been tremendously successful in many fields: our recent ability to learn very flexible functions between input features, x
, and output y
using (extremely) large datasets has profoundly altered our ability to (e.g.,) create predictive models and realize computer vision and speech. However, often, the models we create are used to plan interventions
; in healthcare for example a model mapping features of a person and a treatment might be used to predict survival rates (this is just another model we can learn based on data), and that model is subsequently used to choose the best treatment.
Regretfully, models trained on observational data often do not properly predict outcomes when changes are made to the underlying process. For example; when training a model using the treatment (chemotherapy, radiotherapy, and / or surgery) as input to predict survival rates of breast cancer patients on observational data, one finds that refraining from chemotherapy improves
survival. However, this is largely illusionary; a common cause -- namely the severity of the tumor -- causes both the decision by the specialist to administer chemotherapy and
influences the survival rate. The naïve model trained on observational data is erroneous when considering the causal effect of the treatment.
This remains true irrespective of the amount of data available.
In recent years many advances have been made in the field of estimating causal effects based on observational data. These methods however rely on a large number of assumptions, some of which are unreasonable for many data science applications. In this PhD project we would like the PhD student to a) study and compare different contemporary methods of dealing with causality, b) relax a number of the current assumptions and develop a Bayesian approach to causality, and c) develop -- in combination with clinical practitioners -- data science methods that allow for better estimation of causal effects based on a combination of observational and interventional data.Work environment and supervision
The position will be supervised by prof. dr. Maurits Kaptein (TiU); principal investigator of the computational personalization lab
at JADS (see www.nth-iteration.com
). The work of the lab finds applications both in healthcare and in (online) marketing. The lab consists of the PI, 5 PhD students, and one post-doc.
The computational personalization lab is part of the 'Data & Humans' theme
(headed by dr. Martijn Willemsen & prof. dr. Maurits Kaptein). Within the theme prof. dr. Eric Postma (TiU), prof. dr. Chris Snijders (TU/e), prof dr. Wijnand IJsselsteijn (TU/e), and prof. dr Jack van Wijk (TU/e) will contribute to the project and supervision. Other research labs within the theme focus on understanding recommender systems
, data visualization
, and modeling networked sensor data
The PhD student will be part of the JADS eco-system: both the Data & Humans theme as well as JADS as a whole offer a strong PhD program, frequent scientific seminars by internal and external speakers, and the ability to follow graduate level courses at both TiU and TU/e. JADS offers the unique opportunity to be embedded in a starting research institute, with its associated freedom, while being firmly grounded and supported by experts from two top Dutch Universities.
The positions includes allowances for research expenses, including computer, participant costs, experimental materials, and travel and subsistence costs towards project-specific travel, including conferences. The positions also offers an opportunity to gain experience with teaching, subject to availability and qualifications.