List of Projects – Autumn 2021
Project supervisor- Berfin Simsek, co-supervised with François Ged.
What are we curious about? An empirical study – assigned
Humans not only seek external rewards such as money or food but also like to solve puzzles, try new activities, watch series, and do many other tasks that are not directly related to their primary needs and survival. In neuroscience and psychology, these activities are often interpreted as curiosity-driven activities. In the reinforcement learning (RL) framework, the curiosity-driven decisions of humans are modeled as actions towards maximizing an “internal reward” or “intrinsic motivation”. Seeking such intrinsic motivations can help an artificial agent to efficiently explore its environment (for example in Atari games [1,2]) and can explain human exploratory behavior .
In this project, you will compare the effect of seeking different intrinsic motivations on the performance of an RL agent in different types of environments. We will stick to tabular settings and use classic model-based RL algorithms augmented with intrinsic motivations. The hope is to find better intuitions about the potential answers to “Why are humans curious?” and “What are they curious about?”.
– You should be familiar with RL theory and different RL algorithms; it would be ideal if you have passed the Artificial Neural Network course by Prof. Wulfram Gerstner.
– You need to be able to code efficiently and comfortably in either Python or Julia (preferred) language.
Interested students should send an email with their CV and grades to [email protected].