Descrizione

Titolo del progetto di ricerca: "DSSApple-Sviluppo di un sistema informatico per la determinazione delle malattie di post-raccolta delle mele"<br>
Descrizione dell'attività da svolgere dal/la collaboratore/trice: "Development of a Reinforcement Learning environment for training an agent in order to learn to recommend items, without any prior knowledge. The agent will start by playing episodes or else user sessions (recommending some items to different users) and as the training goes on, the more episodes/user sessions the agent plays, the better it will learn to recommend items. As a RL agent, we will use various RL algorithms such as DQN (and the RAINBOW version of it) which is an off-policy algorithm and the A2C which is an on-policy method. We will compare the accuracy and other metrics of these algorithms on various datasets to other baseline recommendation algorithms (such as the RWR and the Simrank algorithms) in order to find out if it achieves better results. Our goal is to create an agent which will be able to be trained on the fly as new data is obtained and increase its accuracy. All deliverable algorithms in the end should be implemented in Python."<br>
Responsabile: prof. Markus Zanker<br>