@prefix this: . @prefix sub: . @prefix np: . @prefix rdf: . @prefix prov: . @prefix npx: . @prefix dc: . @prefix xsd: . a np:Nanopublication; np:hasAssertion sub:assertion; np:hasProvenance sub:provenance; np:hasPublicationInfo sub:pubinfo; dc:created "2026-05-08T10:54:41.997Z"^^xsd:dateTime; dc:creator ; dc:license ; npx:introduces sub:sphere-a-to-b; npx:wasCreatedAt ; "Sphere matched filter transfers A→B at 1.000 without retraining; flat drops from 1.000 to 0.845 on the same transfer"; . sub:sphere-a-to-b a ; "2026-05-06"^^xsd:date; "Sphere matched filter transfers A→B at 1.000 without retraining; flat drops from 1.000 to 0.845 on the same transfer"; "The substrate-rotation-equivariance property the within-discipline test (chain A) demonstrated also delivers cross-discipline transfer at the same magnitude on the HEALPix-NESTED substrate. The sphere-harmonic band-pass matched filter trained on a cosmology-like domain (uniformly-random feature locations, steep background spectrum) classifies a climate-like domain (high-latitude features, smoother background plus cosine-of-latitude baseline) at 1.000 accuracy without retraining. The equivalent lat-lon flat matched filter trained identically drops from 1.000 in-domain on the cosmology-like domain to 0.845 on the cross-domain transfer to the climate-like domain because the equator-shape template under-responds to polar-stretched features in lat-lon space. Both pipelines reach 1.000 / 0.995 in-domain on the climate-like domain when trained directly on it, so the test is fair — the asymmetry shows up only in the cross-discipline transfer column. This is the operational claim that investments in sphere-aware models from one discipline (astrophysics / cosmology, where the HEALPix ML stack is most mature — DeepSphere, foscat, healpy) carry over to other disciplines (climate, biodiversity, Earth observation) on the shared HEALPix substrate without retraining."; ; """Numerical results from notebook 06 — sphere-aware {A→A in-domain 0.990; A→B transfer 1.000; B→B upper-bound 1.000} versus lat-lon-flat {A→A in-domain 1.000; A→B transfer 0.845; B→B upper-bound 0.995}. 200 training samples + 100 test samples per class per domain. Identical (max, mean, std) features and identical logistic-regression classifier heads on both pipelines. Reproducible end-to-end via the repository's environment.yml + Snakefile. Github repository: https://github.com/annefou/spherical-ml-biodiversity"""; """(i) Synthetic domain regimes constructed to share feature physics across different background spectra; true cross-discipline transfer from real cosmology data (e.g. Planck CMB on HEALPix) to real climate data (e.g. DLWP-HEALPix forecasts on HEALPix) would require integrating with foscat scattering networks or a DeepSphere graph convolutional network as future work. (ii) The substrate effect is isolated from the model class via the minimal (max, mean, std) feature triple; richer learned representations would deliver substantially different absolute accuracy numbers but the substrate-dependence is the geometric mechanism the experiment captures. (iii) The latitude restriction in the climate-like domain is the regime where lat-lon projection distortion bites hardest; the cross-discipline transfer test would yield different numbers for differently-distributed feature regimes. (iv) Two domains tested; the transfer-between-pairs claim generalises naturally to N-way transfer but was not separately tested with three or more domains. """; ; ; . sub:assertion prov:wasAttributedTo . "Anne Fouilloux" . sub:sig npx:hasAlgorithm "RSA"; npx:hasPublicKey "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"; npx:hasSignature "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"; npx:hasSignatureTarget ; npx:signedBy .