SOURCE-LINKED INTELLIGENCE

WaterKron and FlipFlop Hessian: Information-Theoretically Grounded Quantization with Kronecker-factored Hessians

arXiv · AI, language, vision and robotics · article · Sep 13, 2026 · UTC

How should a Kronecker-factored Hessian approximation be chosen for post-training quantization? We address this question through WaterKron, which combines two-sided GPTQ with row- and column-dependent waterfilling scales and entropy coding. We derive its high-rate distortion with respect to the full Hessian using an explicit Kronecker-Hessian mismatch factor $Φ$. This factor quantifies the asymptotic distortion penalty due to the Kronecker Hessian approximation and provides a criterion for selecting the factors optimally. Minimizing $Φ$ leads to a Gaussian covariance-fitting problem with class

Read original source ↗ Open in workspace

recordType

paper

region

Global