Differential privacy support for graph convolutional layer
Motivation
Graphical deep neural networks based has become integral to solving many real-world problems. Their applications will only continue to increase in the coming years. Their ability to integrate spatial and other thing of information allows us to model multiple linked nodes using a single model and also learn the interaction between multiple nodes (Traffic). The additional information gained due to interaction between nodes poses a privacy problem. Providing support to the graphical modeling layers would alleviate this issue to an extent and allows to bring much experimental research to real-world with the concern of privacy
🚀 Feature
Differential privacy support for graph convolutional layer
Motivation
Graphical deep neural networks based has become integral to solving many real-world problems. Their applications will only continue to increase in the coming years. Their ability to integrate spatial and other thing of information allows us to model multiple linked nodes using a single model and also learn the interaction between multiple nodes (Traffic). The additional information gained due to interaction between nodes poses a privacy problem. Providing support to the graphical modeling layers would alleviate this issue to an extent and allows to bring much experimental research to real-world with the concern of privacy
Alternatives
Additional context