Dimension-free comparison inequalities between $f$-divergences for Gaussian location mixtures

概要

We establish dimension-free comparison inequalities between the Kullback-Leibler (KL) and $χ^2$-divergences and the squared Hellinger distance for Gaussian location mixtures with compactly supported mixing distributions. We show that both divergences are bounded by the squared Hellinger distance up to constants depending only on the radius of the support, independently of the ambient dimension and the number of mixture components. These results resolve an open question raised by \cite{jia2023entropic} and strengthen existing comparison inequalities. Our proof relies on a Gaussian interpolation argument that relates the derivatives of these divergences to differences between posterior means.

参考リンク

• Paper ↗

リンク先のライセンス・再現性・データ条件をご確認ください。自動分類や実装の目安は保証ではありません。

実装を始めるプロンプト

参考リンクを確認したうえで、次の内容をコピーして利用できます。

# 論文・記事
Dimension-free comparison inequalities between $f$-divergences for Gaussian location mixtures

# 概要
We establish dimension-free comparison inequalities between the Kullback-Leibler (KL) and $χ^2$-divergences and the squared Hellinger distance for Gaussian location mixtures with compactly supported…

# 原典
https://arxiv.org/abs/2610.25070

上記の手法を実装してください。実行前に依存関係とライセンスを確認してください。