My primary objective for FIRE, a neural network stuff model, was to provide diversity, with the hope that being different will be additive, rather than repetitive. It appears to have achieved that objective, as simply averaging FIRE with StuffPro improves correlations.
7 months ago