|BERNARDIS CESARE||Cycle: XXXIV |
Section: Computer Science and Engineering
Tutor: TANCA LETIZIA
Advisor: CREMONESI PAOLO Major Research topic
:Generative Adversarial Networks for Recommender SystemsAbstract:
Generative Adversarial Network, or GAN, is an unsupervised learning technique presented in 2014 by Ian Goodfellow. It consists in a couple of neural networks that compete one against the other in a zero sum game. The first network, called generative, has to generate candidates that have to be evaluated by the second network, called discriminative. The objective of the latter is to recognize which candidates have been generated by the generative network and which are, instead, real candidates from the original dataset.
In the recent past these networks have become very popular, showing a surprisingly good ability in generating realistic images and 3D models. However, only few experiments have been performed in the recommender system field and their applicability has been only superficially explored.
Our purposeis to exploit Generative Adversarial Networks potential in the recommendations domain over different tasks, from the classical top-n recommendation, to the dataset enrichment and more.
Thanks to GAN's promising potential, we expect to obtain excellent results, with the final goal to outperform the actual state-of-the-art.