Description

A famous computational challenge is the Travelling Salesman Problem, in which a traveller needs to find the shortest route to visit a set of cities. Humans and other animals are very good at finding efficient solutions to practical tasks related to the Traveling Salesman Problem, but we do not know which strategies or algorithms they use to solve the problem. In computer science promising approaches to find an optimal solution include (deep) reinforcement learning [1]. Do humans and other animals use a strategy, similar to a reinforcement learning approach? To address this question, this project involves the comparison of reinforcement learning models with animal behavioural data from a Travelling Salesman task. The animal data is available from an ongoing study in the lab of Prof. Jonas Rose (Faculty of Psychology, RUB), in which pigeons have to visit a series of goal locations in a large arena. The navigational paths taken by the pigeons (Fig. 1) indicate potential strategies and algorithms they use to solve the task. In this project you will implement a reinforcement learning model of the animal behaviour in that task and compare the behaviour of the model with the behaviour of the pigeons to identify the underlying strategy and algorithm used by the pigeons. For the simulation of the task and the reinforcement learning model the CoBeL-RL framework can be used [2].

Prerequisites: Python programming, completed courses on artificial intelligence and/or computational neuroscience

References

[1] Munikoti, S., Agarwal, D., Das, L., Halappanavar, M., & Natarajan, B. (2022). Challenges and opportunities in deep reinforcement learning with graph neural networks: A comprehensive review of algorithms and applications. arXiv preprint arXiv:2206.07922.

Fig. 1: Trajectory of a pigeon in the Traveling Salesman task.