Point-wise losses induce an irreducible bias in time series forecasting governed by sequence length and structural signal-to-noise ratio; DFT/DWT orthogonalization plus a harmonized lp norm reduces it and improves accuracy.
While the performance of many baselines (e.g., MICN, Autoformer) degrades sharply as more data points are masked, the lead of LHarm,ℓp remains remarkably stable
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
The Procrustean Bed of Time Series: The Optimization Bias in Point-wise Loss Functions
Point-wise losses induce an irreducible bias in time series forecasting governed by sequence length and structural signal-to-noise ratio; DFT/DWT orthogonalization plus a harmonized lp norm reduces it and improves accuracy.