A carregar...
Publicação
Training tabular generative adversarial networks in a federated learning framework for generating realistic, non-sensitive data for Modatta: a case study of recommendation systems in hyperbolic space
| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 1.71 MB | Adobe PDF |
Orientador(es)
Resumo(s)
The goal of this work project is to explore the usefulness of a privacy-preserving
framework that enables Modatta to make personalized recommendations to users in their
application. For that purpose, the Federated Training of Generative Adversarial Networks for
Tabular data was studied, and its performance was evaluated on generated synthetic data. The
synthetic generated data from this type of model allowed the training of the Recommendation
System. Choosing the right users for campaigns has a huge impact on Modatta’s user experience
and satisfaction, therefore and due to the hierarchical nature of users interests data, Hyperbolic
Recommendations System models were investigated.
Descrição
Palavras-chave
Data-privacy Generative adversarial networks Federated learning Systems Tabular data Machine learning Deep learning Hyperbolic embeddings Hyperbolic recommendation system Hierarchical data