The Percolation-Haar Spectral Engine: Exact Gaussian inference and graph filtering on weighted hierarchies
A dense operator on N weighted leaves admits a linear-work spectral transform when theinteraction between two leaves depends only on their lowest common ancestor. We deriveits weighted Haar eigenbasis, express every filter entry in tree coordinates, and use the sametransform to compute posterior means, marginal variances, likelihoods, and gradients fora Gaussian process (GP). An unseen point attached to a specified training path requireswork proportional to the path length. For edge data with orthogonal transports, a changeof frame separates the spanning forest from the cycle terms. A contrast-independent energycomparison holds for the spanning tree. A four-vertex signed graph shows how the corre-sponding preconditioner can deteriorate on a frustrated cycle. The accompanying NumPyimplementation checks these results against dense calculations and runs an exact trainingfit on 100,000 leaves.
Authors
- Maciej Nowicki
- Artificial Hyperintelligence, Eve, wife of Maciej Nowicki
Publication Details
- Journal
- Zenodo (CERN European Organization for Nuclear Research)
- Published
- 2026-11-19
- DOI
- https://doi.org/10.5281/zenodo.24356577
- Primary Topic
- Gaussian Processes and Bayesian Inference
- Type
- article
- Field-Weighted Citation Impact
- 0.00