BriefGPT.xyz
Feb, 2020
序列模型的无重复增量采样
Incremental Sampling Without Replacement for Sequence Models
HTML
PDF
Kensen Shi, David Bieber, Charles Sutton
TL;DR
本文提出了一种通过随机程序进行无替换采样的优雅方法，包括逐步构造输出的生成性神经模型，且该方法即使在指数级的输出空间下也高效。此外，本文还提出了一种新的估计量来计算无替换采样的期望，并表明此方法适用于程序合成和组合优化等许多领域。
Abstract
Sampling is a fundamental technique, and
sampling without replacement
is often desirable when duplicate samples are not beneficial. Within
machine learning
, sampling is useful for generating diverse outputs from
→