BriefGPT.xyz
Nov, 2024
简化贝叶斯深度学习中的预测过程
Streamlining Prediction in Bayesian Deep Learning
HTML
PDF
Rui Li, Marcus Klasson, Arno Solin, Martin Trapp
TL;DR
本研究针对贝叶斯深度学习（BDL）中预测计算效率不足的问题，提出了一种通过单次前向传播而无需采样的方法。通过对激活函数的局部线性化和线性层的局部高斯近似，我们能够分析性地计算后验预测分布的近似值，从而提升预测效率。我们的方案在多层感知机和变压器模型（如ViT和GPT-2）上的回归和分类任务中表现良好。
Abstract
The rising interest in
Bayesian Deep Learning
(BDL) has led to a plethora of methods for estimating the posterior distribution. However, efficient computation of inferences, such as predictions, has been largely overlooked with
→