Extend TailProp beyond Gaussian and Cauchy propagation bases

Develop broader stable-process propagation families or learnable propagation basis sets for visual representation learning to capture regimes beyond the Gaussian and Cauchy bases used by TailProp.

Background

TailProp combines two complementary stable-process propagators: light-tailed Gaussian propagation and heavy-tailed Cauchy propagation. The authors explicitly restrict their study to these two bases and identify the extension to broader stable families or learnable collections of propagation bases as unresolved. Such extensions could provide a richer set of spatial interaction and decay profiles for heterogeneous visual representations.