MetricOpt: Learning to Optimize Black-Box Evaluation Metrics
Backpropagate through your non-differentiable metric function
#optimization
#loss-function
#gradient-descent
Gradient Matching for Domain Generalization
Guiding neural networks to generalize across unseen data domains
#generalization
#loss-function
#gradient-descent
Orthogonal Projection Loss
Simple method with simple theoretical background to improve model performance supported by extensive experiments
#loss-function
#orthogonality