Michael M. Bronstein, Joan Bruna, Taco Cohen, Petar Veličković (2021), arXiv preprint (proto-book) — ARXIV-2104.13478
Standing in the record
Registered on 2026-09-29 at the owner’s request, because the owner’s prior-art novelty map (a private working document on concepts as projectors, the linear representation hypothesis and their neighbours) cites it: row 7: geometric deep learning / equivariance (impose symmetry vs. the response’s detect-symmetry). It is seeded from metadata and published descriptions and has not been read here.
Identification notes: Map gives no year; the 2021 arXiv proto-book is the canonical “Bronstein et al.” GDL reference. Alternative: Bronstein, Bruna, LeCun, Szlam & Vandergheynst 2017, “Geometric deep learning: going beyond Euclidean data”, IEEE Signal Processing Magazine (doi 10.1109/MSP.2017.2693418) — not registered.
It was filed Deferred, unread. NOTE-274 is the close reading of 2026-09-29, and it placed the work: Active — worth reading as the standard systematisation of symmetry-as-architecture in deep learning and for its clean derivation of convolution and the Fourier basis from translation symmetry (§4.2). Read it as a proto-book that defers its representation theory (irreps, group Fourier transforms) and its learning theory to “future work”. It takes the symmetry group as known and never addresses discovering one.