A Non-Parametric Test to Detect Data-Copying in Generative Models

A Non-Parametric Test to Detect Data-Copying in Generative Models
复制标题

DOI:
--
复制
发表时间:
2020-04
期刊:
ArXiv
影响因子:
--
通讯作者:
Casey Meehan;Kamalika Chaudhuri;S. Dasgupta
Casey Meehan;Kamalika Chaudhuri;S. Dasgupta
中科院分区:
其他
文献类型:
--
作者:
Casey Meehan;Kamalika Chaudhuri;S. Dasgupta

文献摘要

相似文献

在机器学习中,检测生成模型中的过拟合是一项重要挑战。在这项工作中,我们将一种过拟合形式形式化,我们称之为“数据复制”——即生成模型记住并输出训练样本或其微小变体。我们提供了一种用于检测数据复制的三个样本非参数检验,该检验使用训练集、来自目标分布的一个单独样本以及从模型生成的一个样本,并研究了我们的检验在几个经典模型和数据集上的性能。如需代码和示例,请访问此https链接
Detecting overfitting in generative models is an important challenge in machine learning. In this work, we formalize a form of overfitting that we call {\em{data-copying}} -- where the generative model memorizes and outputs training samples or small variations thereof. We provide a three sample non-parametric test for detecting data-copying that uses the training set, a separate sample from the target distribution, and a generated sample from the model, and study the performance of our test on several canonical models and datasets. For code \& examples, visit this https URL