Revisiting Long-term Time Series Forecasting: An Investigation on Linear Mapping

In three linesarXiv study on long-term time series forecasting (LTSF). Authors show that a simple linear layer (affine mapping) dominates performance on standard benchmarks. Analysis reveals models learn similar transition matrices, capture periodic patterns well but fail on non-periodic signals. Code available.

Summary generated by Claude — human-verified