MorphStrata: Layer-Specific Perturbations for Generating Morphence Students in Time-Series Moving Target Defense

In three linesMorphStrata enhances Moving Target Defense for time-series forecasting models via selective layer-specific stochastic noise injection. Tested on Transformer with FGSM, BIM and PGD attacks, the approach reduces adversarial RMSE by up to 97.97% on AEP data with training overhead <1%.

Summary generated by Claude — human-verified