Dataset

Metrics

Methods

1. Data Augmentation.

2. Adjust Loss Function.

   • Adding Penalty Terms (L1 Norm, L2 term).

   • Weight Decay (w_new = w_old - η * (∇L + 2λw_old)) with ∇L is the gradient of the original loss function.

3. Noise Injection.

4. Mixup.

5. Larger Learning rate.

6. Smaller Batches.

7. Neural-network Specific.

   • Dropout.

   • Batch Normalization (BatchNorm)

8. Early Stopping.

9. Multi-task Learning.

10. Ensembles.

11. Label Smoothing.

Paper List