Mechanism diagram · MLP / Linear
MLP / LINEAR Input lookback L flat features Dense layer with activation Dense layer + residual Linear head Linear floors (DLinear / NLinear-RevIN) collapse to one dense layer.

How it works

The flattened lookback window is fed to a sequence of dense layers with non-linear activations.

Final layer outputs the forecast horizon directly.

No structural priors — relies entirely on capacity and regularisation.

Pros and cons on this universe

Pros

  • Cheapest possible incumbent — fast to train.
  • CPCV history available for comparison against more elaborate arms.

Cons / failure modes

  • Sub-zero IC (−0.009) on this universe.
  • Highly scaling-sensitive — small changes in normalisation can flip the sign of performance.
  • Beaten by every structured architecture on the screen.

References