摘要
生成模型是机器学习中一类重要的模型,它们通过学习数据中的统计规律,能够生成具有相似特征的新样本。然而,传统生成模型对于因果关系的建模能力较弱。随着机器学习领域的发展,越来越多的研究者试图从因果关系中分离出可能的虚假相关性,基于因果推断的生成模型在这一背景下值得被研究。现通过对生成模型、因果生成模型的研究进展做出阐述,介绍其特点及代表性方法。相比之下,因果生成模型展现出了较优的性能和较好的可解释性。
Generative models are a crucial class of models in machine learning.By learning statistical patterns from data,they can generate new samples with similar features.However,traditional generative models have limited capabilities in modeling causal relationships.As the field of machine learning evolves,an increasing number of researchers are attempting to disentangle potential spurious correlations from causal relationships.In this context,generative models based on causal inference deserve inves-tigation.This article provides an exposition of the research progress in generative models and causal generative models,outlining their characteristics and representative approaches.In comparison,causal generative models demonstrate superior performance and enhanced interpretability.
作者
李晓晋
刘进锋
Li Xiaojin;Liu Jinfeng(School of Information Engineering,Ningxia University,Yinchuan 750021,China)
出处
《现代计算机》
2023年第23期37-41,共5页
Modern Computer
基金
宁夏自然科学基金项目(2023AAC03126)。
关键词
生成模型
因果推断
可解释性
虚假相关性
generative models
causal inference
interpretability
spurious correlations
