Anchor Data Augmentation (ADA) enhances robustness in regression tasks by combining samples based on clustering, outperforming many existing augmentation techniques in limited data scenarios.
Through ADA, we systematically mix samples according to their collective similarities via clustering, pushing modifications toward or away from centroids to achieve targeted robustness,
Empirical evaluations across a range of synthetic and real-world regression problems reveal that ADA not only competes but often surpasses state-of-the-art data augmentation methods.
The flexibility of ADA allows it to be applicable to various regression problems, demonstrating consistent improvements in performance without detrimental effects on any dataset.
Collection
[
|
...
]