FedTLBOHB: Efficient Hyperband with Transfer Learning for Vertical Federated Learning

[nabenabe0928]

FedTLBOHB: Efficient Hyperband with Transfer Learning for Vertical Federated Learning

Low quality.

Federated HPO ==> we have various HPO results from the same model with different datasets

Main points

• Use GP(?)-based BO, but not TPE but anyways BOHB
• Learn the weights of each surrogate and the weighting algorithm is inspired by AdaBoost although I am really not sure the regression version of AdaBoost works well for the surrogate models on different $p(y)$