pythonpapertime-seriesnbeatsforecastingafter-work
Introducing N-BEATS: Neural basis expansion analysis for interpretable time series forecasting for after-work quants
Oreshkin, Carpov, Chapados and Bengio (ICLR 2020; arXiv:1905.10437) stack fully-connected residual blocks with a forecast/backcast fork. No TS feature engineering. Generic N-BEATS beats the M4 hybrid winner by ~3% OWA; an interpretable trend+Fourier variant stays within a hair. M3 and tourism too. Plain-language intro. Not investment advice.