ConceptDimensionless Prior
Dimensionless Prior
A network design inspired by dimensional analysis in physics that makes the model predict only the normalized direction field or shape pattern of the residual while ignoring absolute scale, thereby improving cross-domain generalization.
Timeline (last 90 days)
Oct 14
Cepheid's dimensionless prior divides the residual by its own root-mean-square amplitude, bringing the output near unit scale, so that the residual spatial distribution patterns of different families fall within the same numerical range.
Unverified50%