AI RESEARCH
Mechanistic Evidence for Spectral Structures in Prior-Data Fitted Networks
arXiv CS.LG
•
ArXi:2601.21731v2 Announce Type: replace Prior-Data Fitted Networks (PFNs) enable amortized Bayesian inference in a single forward pass, yet their internal representations remain opaque. It is unknown whether PFNs encode identifiable Bayesian structure or merely memorize input-output mappings. We provide mechanistic evidence that PFNs learn structured spectral representations and that these can be extracted as explicit kernels.