Artificial Intelligence (COMPSCI_4004) coursework 2021

The report investiaged the performance of three agents - random, deterministic and Q-Learning agent - and compared the performance of these agents against each other using three different simulated epidemic environment conditions. Utilising previous evaluation results, the report looked at possible improvements to the Q-Learning agent to perform better in the stochastic environment. It was found that the Q-Learning agent performed the best and produce the most sensible policies, although in a real life instance none of the policies produced by the agents could be fully used to inform decisions due to the limitations in the environment and other social factors involved.

ViRL is an Epidemics Reinforcement Learning Environment. Agents are tasked with controlling the spread of a virus with one of four non-medical policy interventions: (i) lockdown, (ii)track & trace, (iii) social distancing & masks, (iv) none.

Sebastian Stein / ViRL GitLab. Available at: https://git.dcs.gla.ac.uk/SebastianStein/virl