作者:禅与计算机程序设计艺术
1.简介
“知识转移”这个概念从古至今都存在着,但在最近几年开始被越来越多地应用到机器学习领域中。它可以理解为利用已有的知识进行预测或者分类,从而帮助提高模型的性能。然而,传统的基于规则或统计的方法往往存在着偏差,因此如何将已有的知识迁移到新的领域、场景下仍然是一个重要的研究方向。近些年来,神经网络(Neural Network)被广泛用于解决这一问题。其中Adversarial Autoencoder (AAE)是一种最成功的知识迁移方法之一。本文将简要介绍AAE的相关知识,并通过一些具体实例来阐述其原理和运用方式。希望对读者有所启发!
2.相关论文及引用
- Learning Deep Representations by Generative Adversarial Networks[1]
- Transferring knowledge from trained models to novel domains through adversarial autoencoders[2]
- Unsupervised Domain Adaptation with Generative Adversarial Networks[3]
[1], . “Learning deep representations by generative adversarial networks.” arXiv preprint arXiv:1511.06434 (2015).
[2], et al. “Transferring knowledge from trained models to novel domains through adversarial autoencoders.” arXiv preprint arXiv:1907.06312 (2019).
[3], et al. “Unsupervised domain adapt