ConceptAsymmetric Bernoulli Perturbation
Bernoulli Perturbation Augmentation
A data augmentation method for binary features that flips the 0/1 value of feature bits with asymmetric probabilities to generate structure-preserving sample variants. It is commonly used with binary data such as symptom diagnosis to improve model regularization and generalization.
Timeline (last 90 days)
Nov 2
The experimental design used a 2×2 matrix that fully crossed the two factors of Bernoulli perturbation augmentation and hyperparameter tuning, in order to separate the contributions of augmentation and tuning.
Unverified50%