Mixup: Blending the Labels
Beginner
The problem, from first principles
Mixup mixes labels exactly as it mixes inputs, so a blended example gets a soft target. The model is then trained to match that soft label, not a hard one.
From theory to code
Implement mixup_targets(y, perm, lam), applying the same permutation and weight to the one-hot labels.
Constraints
- Each output row sums to 1 when the input rows do.
Hints
Hint 1
Use the same expression as the inputs, applied to the label matrix.
Keywords
research-papersaugmentationmixup
Python
Run your code to see results
No results yet
Run your code to check it against the examples, or submit to run every test.