On di-Cayley graphs and their spectrum
arXiv:2608.27231
2026
Architecture
1 ideas extracted · analyzed Aug 29, 2026
What the math gives to ML
The paper's transferable asset is an explicit representation-theoretic decomposition of a two-part directed graph into independent spectral blocks. For each irreducible representation, within-stream and cross-stream connection sets become small matrices, with the displayed block M_ρ = [[ρ(S), ρ(T)], [ρ(T)*, ρ(S)]]. This suggests a two-stream group-equivariant neural layer whose spatial mixing is performed independently in Fourier modes, reducing structured mixing from quadratic cost to FFT-based cost for cyclic groups. The first implementation should target periodic sequences, grids, or graph data with a known finite-group action, where exact equivariance is useful and the structural assumption is falsifiable.
Ideas from this paper
Unverified
Re-invented
2026
Represent tokens or nodes as two feature streams indexed by a finite group G, and replace dense pairwise mixing by a learned di-Cayley operator with within-stream and cross-stream connections. In the Fourier or irreducible-representation basis, each group frequency is processed independently by a small block matrix, giving exact translation or group equivariance and O(|G| log |G|) mixing for cyclic groups.
Useful6/10
Difficulty5/10
Novelty5/10