Model

Model architecture

Experiment

Train 3 models

• MegaNegRaising

• MegaVeridicality

• Both combined

Isotonic regression

• Four-fold cross validation on pilot data

• Fit isotonic regression to calibrate model predictions

Other details

• BlueHive cluster of Intel Xeon CPUs

• 10 epochs of training

• MSE loss, Adam optimizer, batch size of 1

• Dropout