Moonfish en conférence à Bangkok

Le projet MOONFISH était en Thaïlande à Bangkok pour la conférence OLA’2019 «International Conference on Optimization and Learning». Elle s’est tenue du 29 au 31 janvier et portait sur l’optimisation et l’apprentissage automatique.

Notre doctorant, Nicolas Poiron-Guidoni, a présenté ses travaux sur la différence entre l’apprentissage par renforcement et les méthodes d’optimisation, appliqués à la gestion de la pêche corse.

Le titre et le résumé de sa présentation sont les suivants :

Comparison between reinforcement learning and optimization methods applied to the management of the corsican multi-species fishery

In this article, we present two decision support approaches applied on a fishery model. The first consists about training a neural network via Reinforcement Learning to determine a real time optimal fishing policy. The second uses optimization via simulation methods to determine optimal fishing quotas. The aim of our work is the simulation of biological model based on a dataset, collected and completed via simulated data, to make a decision support tool to help every fishermen in their job and allow stocks restoration for endangered species. We propose recommendation of fishing areas for each simulation time and then, study the fishing effort distribution during a year taking into account more complexes parameters like price variations. But the reinforcement learning approach is new in fishery management, so it is important to first compare the results with an improved one to determine the advantages of each one. Finally, we propose amelioration perspectives.

Le lien de la conférence : https://ola2019.sciencesconf.org/