Dual-headed SplitNN #46
Description
Feature Description
Implement a dual-headed splitNN. Each head takes some data as input and computes some representation of the data. The two intermediate vectors are combined and the rest of the network computes on the combined data. We should apply this network to Synthea medical data (#40). We need to split the data in a way which makes sense in a real setting.
When training this model, don't worry about the PSI process to link data entities. We will remove/jumble datasets to build the story at a later stage.
Is your feature request related to a problem?
In many real-world situations, input data is split across data holders. The current implementation of SplitNN takes input data from only one source, and the other data holder is expected to hold the labels.