. . . . "Lat-lon-flat machine learning trained identically on the first discipline's data and applied to the second discipline's data via the equirectangular projection, exploiting only translation equivariance in pixel space." . "Sphere-aware machine learning trained on one discipline's data and applied to another discipline's data on the shared HEALPix substrate without retraining, exploiting rotation equivariance on the sphere." . "Detection accuracy on the second discipline's test set without retraining (cross-domain transfer accuracy), with in-domain accuracy on each discipline as sanity check and upper bound." . "Pairs of scientific disciplines whose data live on the sphere — for example cosmology / astrophysics, climate, Earth observation, marine biodiversity — with shared HEALPix substrate but discipline-specific background spectra and feature-location distributions." . . . . . "For two scientific discipline regimes that share the HEALPix substrate but differ in their background statistics and feature-location distributions, does a sphere-aware machine-learning approach trained on one discipline classify the other discipline without retraining at higher accuracy than an equivalent lat-lon-flat machine-learning approach trained identically, measured by cross-domain transfer accuracy?" . . . "Does sphere-aware ML on HEALPix transfer cleanly across discipline pairs without retraining?" . . "Anne Fouilloux" . "2026-05-07T21:09:34.069Z"^^ . . . . . . . . . "RSA" . "MIGfMA0GCSqGSIb3DQEBAQUAA4GNADCBiQKBgQDWv2pJnmDsBOq8OlT1aSvYXSuWT34WOp4FYqEzdnn2F0kqzcFevBqWGZDxJWC0lqCrDEuNfp2QFyPe/+nES9dlHGYIhqPi68fwK6ZiNUotRFxXou+rjFznVvUxtCL8Ede79EBHwWN61QtwSIcU12bLoZsNPFlqQASQ93BJuKlihwIDAQAB" . "gbTYw+4diAm0p3FXbRcY8KYf1yMaxdE2C6YrpKM7Jh3hJZrGhXLaX0Fo7uVqNn7FVLj5TSwlnet1oYaXeuTVEe+AxACg67Cf64ZyC262OXScPD/M2vQvRcI1ZLHCS5XxqMt+jElQHehc09ZeIprrGYwBUlhQ7kyCIY0r9S8eXkU=" . . .