转发：转发: 周六讨论班

[黄孝鹏]

	黄孝鹏 学生
中国科学院数学与系统科学研究院 邮箱：huangxia...@126.com

签名由 网易邮箱大师 定制

--------- 转发邮件信息 ---------

发件人： Zhang Chihao <zhangc...@outlook.com>

发送日期： 2018年9月7日 18:10

发送至： 黄 孝鹏 <huangxia...@126.com>

主题： 转发: 周六讨论班

---

发件人: Zhang Chihao <zhangc...@outlook.com>
发送时间: 2018年9月7日 15:17
收件人: zhangrou...@googlegroups.com
主题: 周六讨论班

 

大家好，

周六我将为大家介绍一篇基于GAN的非监督学习方法，标题为：Controllable Invariance through Adversarial Feature Learning

以下为文章摘要：

Learning meaningful representations that maintain the content necessary for a particular task while filtering away detrimental variations is a problem of great interest in machine learning. In this paper, we tackle the problem of learning representations invariant to a specific factor or trait of data. The representation learning process is formulated as an adversarial minimax game. We analyze the optimal equilibrium of such a game and find that it amounts to maximizing the uncertainty of inferring the detrimental factor given the representation while maximizing the certainty of making task-specific predictions. On three benchmark tasks, namely fair and bias-free classification, language-independent generation, and lighting-independent image classification, we show that the proposed framework induces an invariant representation, and leads to better generalization evidenced by the improved performance.

欢迎大家参加。

祝好，

驰浩