favae Sequence Disentanglement using Information Bottleneck

FAVAE: Sequence Disentanglement using Information Bottleneck Principle

FAVAE: Sequence Disentanglement using Information Bottleneck Principle

https://github.com/favae/favae_ijcai2019 效果非常棒

https://arxiv.org/pdf/1902.08341.pdf

Abstract

We propose the factorized action variational au- toencoder (FAVAE), a state-of-the-art generative model for learning disentangled and interpretable representations from sequential data via the infor- mation bottleneck without supervision. The pur- pose of disentangled representation learning is to obtain interpretable and transferable representa- tions from data. We focused on the disentangled representation of sequential data since there is a wide range of potential applications if disentangle- ment representation is extended to sequential data such as video, speech, and stock market. Sequential data are characterized by dynamic and static fac- tors: dynamic factors are time dependent, and static factors are independent of time. Previous models disentangle static and dynamic factors by explic- itly modeling the priors of latent variables to distin- guish between these factors. However, these mod- els cannot disentangle representations between dy- namic factors, such as disentangling ”picking up” and ”throwing” in robotic tasks. FAVAE can dis- entangle multiple dynamic factors. Since it does not require modeling priors, it can disentangle ”be- tween” dynamic factors. We conducted experi- ments to show that FAVAE can extract disentangled dynamic factors 

https://arxiv.org/pdf/1902.08341.pdf

Thus, disentangled representation learning for sequential data opens the door to new areas of research. 

srnn的区别

如果你有志研究和开发自动驾驶技术、提升AI的智能、提高自动驾驶的智能水平，我们欢迎你的加入！

年薪百万来奋斗-骥智CreateAMind2019招聘目标：年薪百万招聘大牛50+  推荐成功送mate20