arXiv:2506.07687stat.MLcs.LG2025-06NeurIPS被引 1

提出更优的梯度估计方法,提升含高斯隐变量模型的训练效果。

Rao-Blackwellised Reparameterisation Gradients

  • 通过罗氏-布莱克韦尔化改进重参数化梯度,降低方差。
  • 在多层贝叶斯神经网络中,初始训练性能显著优于传统方法。
  • 适用于多种需重复使用重参数化技巧的生成模型,尤其适合深度概率建模。

潜变量高斯分布已在概率机器学习中广泛应用。相应地,梯度估计器是实现含潜变量高斯模型梯度优化的核心工具。重参数化技巧常作为默认估计器,因其易于实现且在变分推断中可产生低方差梯度。本文提出R2-G2估计器,即对重参数化梯度估计器进行罗氏-布莱克韦尔化。有趣的是,我们证明贝叶斯MLP的局部重参数化梯度估计器正是R2-G2的一种实例。这一发现使罗氏-布莱克韦尔化梯度的优势可扩展至一系列概率模型。实验表明,在多次应用重参数化技巧的模型中,使用R2-G2进行初始训练能持续获得更优性能。

原文摘要 · Abstract (English)

Latent Gaussian variables have been popularised in probabilistic machine learning. In turn, gradient estimators are the machinery that facilitates gradient-based optimisation for models with latent Gaussian variables. The reparameterisation trick is often used as the default estimator as it is simple to implement and yields low-variance gradients for variational inference. In this work, we propose the R2-G2 estimator as the Rao-Blackwellisation of the reparameterisation gradient estimator. Interestingly, we show that the local reparameterisation gradient estimator for Bayesian MLPs is an instance of the R2-G2 estimator and Rao-Blackwellisation. This lets us extend benefits of Rao-Blackwellised gradients to a suite of probabilistic models. We show that initial training with R2-G2 consistently yields better performance in models with multiple applications of the reparameterisation trick.

梯度估计贝叶斯神经网络变分推断概率建模

Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品;中文卡片由大模型生成,请以原文为准。