Limiting eigenvalue distribution and entropy of multi-Toeplitz matrices
arXiv:2608.17859
2026
Architecture
1 ideas extracted · analyzed Sep 1, 2026
What the math gives to ML
The paper develops a noncommutative analogue of Szegő’s theorem for operators indexed by words in a free semigroup, where finite truncations are naturally d-ary trees rather than one-dimensional sequences. This provides a principled way to build weight-shared tree operators whose parameter count depends on interaction depth instead of the number of nodes. The most transferable construction is a free-semigroup Toeplitz layer, augmented with normalized spectral-distribution diagnostics or regularization to control amplification across tree depth.
Ideas from this paper
Unverified
2026
Represent hierarchical or tree-structured hidden states on words over d symbols and replace a dense mixing layer by a noncommutative Toeplitz operator composed of shared word shifts. Coefficients are reused at every tree location, so the parameter count depends on maximum interaction depth rather than the number of nodes; an optional spectral penalty controls the amplification profile of finite-depth truncations.
Useful6/10
Difficulty5/10
Novelty7/10