Mixup

Mixup: Data augmentation creating synthetic examples by mixing pairs of training samples. Linearly interpolates between two samples and their labels: x = λx₁ + (1-λ)x₂. Improves generalization and calibration. Simple yet effective regularization. Examples include Image classification augmentation, Regularization technique, Improved generalization.

In italiano: MixupLegge III · Come una rete impara dai dati

Linearly interpolates between two samples and their labels: x = λx₁ + (1-λ)x₂. Improves generalization and calibration. Simple yet effective regularization.

Examples

  • Image classification augmentation
  • Regularization technique
  • Improved generalization