Papers of the day   All papers

Flows for simultaneous manifold learning and density estimation

Comments

Kyle Cranmer: Johann & I released v2 of our paper "Flows for simultaneous manifold learning and density estimation" with more experiments. We dubbed the class of model ℳ-Flows Here you see the flow learning the 2d manifold and the density for the Lorenz attractor! 1/n https://arxiv.org/abs/2003.13913 https://t.co/UrR0IF2tAB

7 replies, 704 likes


Kyle Cranmer: Happy to announce our most recent work "Flows for simultaneous manifold learning and density estimation" led by my friend and colleague Johann Brehmer Paper: https://arxiv.org/abs/2003.13913 Code: https://github.com/johannbrehmer/manifold-flow https://t.co/5eGeSiXyey

2 replies, 368 likes


Kyle Cranmer: Accepted! #NeurIPS2020

1 replies, 133 likes


Danilo J. Rezende: Very nice work on learning simultaneously the data manifold and a normalised density on it.

0 replies, 73 likes


Danilo J. Rezende: Very nice paper!

1 replies, 49 likes


Conor Durkan: This paper addresses a major drawback of likelihood-based generative models, which traditionally assume a distribution with full support in the ambient data space, whereas we believe this is likely not the case for highly-structured data like e.g. images. Very exciting work.

1 replies, 40 likes


Daisuke Okanohara: MFMFs are new generative models for data with a low-dimensional manifold, which can provide manifold and density on the manifold. Manifold learning is achieved by AE training, while density estimation is performed by flow-based models on the manifold. https://arxiv.org/abs/2003.13913

0 replies, 13 likes


Olivier Grisel: Interesting paper that pushes for an strict interpretation of the manifold hypothesis of Y. Bengio. Chained invertible neural networks & clever training scheme alternating between reconstruction error to learn the manifold chart & MLE-based density estimation on the manifold.

0 replies, 11 likes


Kyle Cranmer: @MilesCranmer I didn’t mention it in discussion, but this recent work of simultaneously learning a data manifold and a density on it is relevant if one sees the manifold as being defined by imposing symmetry or Noether charge https://twitter.com/kylecranmer/status/1245152269156388864?s=21

1 replies, 7 likes


Kyle Cranmer: Dear procrastination twitter, please vote for the paper thumbnail for our #NeurIPS2020 paper “Flows for simultaneous manifold learning and density estimation” Paper: https://arxiv.org/abs/2003.13913 Choices to follow...

7 replies, 5 likes


Van Bettauer: Impressive work ! Can't wait to try this out

0 replies, 4 likes


Kyle Cranmer: Johann made this nice animation for our recent paper "Flows for simultaneous manifold learning and density estimation" https://arxiv.org/abs/2003.13913 https://t.co/DULprtbxdd

0 replies, 3 likes


HotComputerScience: Most popular computer science paper of the day: "Flows for simultaneous manifold learning and density estimation" https://hotcomputerscience.com/paper/flows-for-simultaneous-manifold-learning-and-density-estimation https://twitter.com/KyleCranmer/status/1268370127705309184

0 replies, 3 likes


Popular ML resources: The most popular ArXiv tweet in the last 24h: https://twitter.com/KyleCranmer/status/1268370127705309184

0 replies, 1 likes


Content

Found on Jun 04 2020 at https://arxiv.org/pdf/2003.13913.pdf

PDF content of a computer science paper: Flows for simultaneous manifold learning and density estimation