Deep hierarchical reinforcement learning in a markov game applied to fishery management decision making
Poiron-Guidoni Nicolas, Bisgambiglia Paul-Antoine · 2020
In this article, we propose an approach based on deep hierarchical reinforcement learning to help the management of a fishery formalized by a markov game. Two actors, the fishermen seeking to maximize their profit and the decision-makers seeking to manage the exploitation in the best possible way through the introduction of quotas. For each, we test two learning approaches, one based only on observations, the other on pre-processing leading to beliefs. Several learning algorithms are tested and the results are compared. This has allowed us to identify a minimum amount of knowledge to have for optimal quota setting that also leads to significant and sustainable gains without prior knowledge.