ARTFEED — Contemporary Art Intelligence

AI-Assisted Proof of Subgaussianity for Quantized Linear Maps

ai-technology · 2026-05-28

A recently uncovered mathematical proof, aided by Gemini 3.5 Flash, demonstrates a subgaussian concentration bound that is independent of dimensions for Gaussian vectors subjected to nonlinear mappings on a coordinate basis. This finding is relevant for any bounded function with a well-conditioned covariance and addresses a query from Simone Bombari concerning sign-quantized linear maps represented as Y = sgn(Wx). This work has been made available on arXiv in the Mathematics > Probability section.

Key facts

  • The result is a dimension-independent subgaussian concentration bound.
  • The bound applies to Gaussian vectors under coordinate-wise nonlinear mappings.
  • The discovery was assisted by Gemini 3.5 Flash.
  • The result applies to any bounded function under a well-conditioned covariance.
  • The tool answers a question by Simone Bombari on sign-quantized linear maps.
  • The note is published on arXiv.
  • The arXiv ID is 2605.27563.
  • The paper is categorized under Mathematics > Probability.

Entities

Institutions

  • arXiv

Sources