Gjbernice

GradNorm: Adaptive Loss Balancing in Multitask Learning

GradNorm addresses the dominating gradients problem in multitask learning by dynamically adjusting task-specific loss weights based on…