A learned mechanism that selectively applies corrections to predictions based on per-dimension scaling factors.