outlier

Proposes Cascade-KDE, a training-free framework that uses two-dimensional kernel density estimation and truncated expectation to restore time-series corrupted by out-of-distribution impulse outliers while preserving local structure and derivative features.