Responsible co-supervisor: Sara Ershadmanesh
Formal supervisor: Peter Dayan
Supervising Collaborator: Dominik Bach (Bonn)
In dangerous or unpredictable environments, humans must decide whether to keep collecting rewards (e.g., food) or to seek information that helps them stay safe. In this project, you will analyse how people plan under threat and how they trade off explicit reward against information-seeking for safety.
We have build a scenario in which biological agents forage for food and must stay clear of threats. We have also collected data from humans performing this task in virtual reality. The goal of the project is to develop computational models based on Partially Observable Markov Decision Processes (POMDPs). These models characterize belief formation about a threat‘s distance and speed, and how these beliefs guide planning in this situation. We will study both optimal behavior and heuristics computational strategies that may explain individual differences in this experiment.
What is available
- Behavioral dataset from a VR experiment
- Collaboration with researchers at the University of Bonn (VR expertise and behavioral analyses)
- A model description and initial codebase
- The active and rich environment at the Computational Neuroscience lab (supervised by Peter Dayan), where you are expected to participate in group activities as well as pursue your Master’s project.
Student tasks
- Improve and clean existing code
- Extend POMDP model to generate optimal behavior
- Compare model predictions to human behavior
- Write the Master thesis
Qualifications/Experience
- Strong programming skills are essential (Python or R)
- Knowledge of POMDPs and experience implementing related models (more experience is a plus)
How to apply
If you are interested and feel confident about this project, please email the following to sara.ershadmanesh@tuebingen.mpg.de by February 10th, 2026:
- CV
- Short description of your experience with POMDPs and computational modeling
- Names and email addresses of two referees
Â