Exceptional supersphere integration and logarithmic Pizzetti kernels
arXiv:2607.21241
2026
Architecture
2 ideas extracted · analyzed Aug 30, 2026
What the math gives to ML
The paper gives a constructive way to handle bilinear structures that become singular at exceptional parameter values: analytically continue the inverse kernel, separate its residue from its logarithmic finite part, and assign each component to the appropriate quotient or radical. The transferable asset is not superspace geometry itself, but a principled replacement for unstable inverse Gram matrices when a polynomial or feature pairing loses rank. In neural networks this suggests kernelized interaction layers and radical-aware regularizers that explicitly separate learnable quotient information from null directions instead of relying on ad hoc pseudoinverses. The approach is most appropriate for polynomial feature maps, equivariant modules, or architectures whose interaction kernels have a tunable formal dimension.
Ideas from this paper
Unverified
2026
When a structured polynomial feature pairing is degenerate, train separately on its nondegenerate quotient and on the explicitly characterized radical instead of allowing both to compete in one singular loss. The quotient branch captures identifiable information, while a transported radical branch preserves information that the ordinary pairing cannot see.
Useful5/10
Difficulty6/10
Novelty9/10
Unverified
2026
Replace a singular inverse interaction kernel by the finite part of its meromorphic continuation at an exceptional dimension, producing an explicit polynomial-times-logarithm feature interaction. This gives a controlled alternative to adding an arbitrary ridge term when a learned polynomial Gram matrix becomes rank-deficient.
Useful5/10
Difficulty5/10
Novelty8/10