Adversarial Training Ineffective on Compressed Datasets

Compressed condensation datasets lose adversarial robustness; new Minimal Finite Covering restores robust accuracy on benchmark image classification. Read more: getnews.me/adversarial-training-ine... #adversarialtraining #datasetcondensation

New Discrepancy Framework Expands Dataset Condensation for AI

A new discrepancy-based framework lets synthetic data sets be far smaller than the original (M ≪ N) while preserving performance and adding robustness and privacy benefits. Read more: getnews.me/new-discrepancy-framewor... #datasetcondensation #discrepancy