Papers of the day   All papers

Generalizing Convolutional Neural Networks for Equivariance to Lie Groups on Arbitrary Continuous Data

Comments

Andrew Gordon Wilson: Translation equivariance on images gives CNNs key generalization abilities. Our new paper "Generalizing Convolutional Neural Networks for Equivariance to Lie Groups on Arbitrary Continuous Data": https://arxiv.org/abs/2002.12880. With @m_finzi, @sam_d_stanton, @Pavel_Izmailov. 1/6 https://t.co/E3LbVcDoZl

9 replies, 404 likes


Simone Scardapane: *Generalizing CNNs for Equivariance to Lie Groups on Arbitrary Continuous Data* [by @m_finzi @sam_d_stanton @Pavel_Izmailov @andrewgwils] Fantastic paper on building convolutional nets that are equivariant to a wide range of transformations. https://arxiv.org/abs/2002.12880 https://t.co/HiRU1plTW7

0 replies, 60 likes


Erik Bekkers: Very interesting work on Lie group equivariant NNs! Some aspects I really like: a way to handle inputs on some space X that is not a homogenous space of G (not done before! G could be smaller than X) and parameterizing kernels with MLPs. Here I give my analysis of the paper. 1/7

1 replies, 43 likes


HotComputerScience: Most popular computer science paper of the day: "Generalizing Convolutional Neural Networks for Equivariance to Lie Groups on Arbitrary Continuous Data" https://hotcomputerscience.com/paper/generalizing-convolutional-neural-networks-for-equivariance-to-lie-groups-on-arbitrary-continuous-data https://twitter.com/andrewgwils/status/1234294393718427648

0 replies, 43 likes


Sam Stanton: How do you use the same neural net architecture to learn equivariant models for molecules, physical systems, and images? Find out in our new paper! https://arxiv.org/abs/2002.12880

0 replies, 23 likes


Maurice Weiler: This is a neat variation of group equivariant convolutions which seems easily applicable to a range of applications beyond image processing. The method is independent from the symmetry group and sampling grids.

1 replies, 16 likes


Kyle Cranmer: Check it out 👇

0 replies, 11 likes


andrea panizza: Year of the Lie Group equivariance! @erikjbekkers @TacoCohen @maurice_weiler @_gabrielecesa_ your citations are on the rise 😀

0 replies, 10 likes


Daisuke Okanohara: We can make convolution layers equivariant for any transformations from Lie group as long as it supports group exp/log maps. Data can be located in arbitrary continuous positions/states (e.g., point cloud, molecules, dynamical system). https://arxiv.org/abs/2002.12880

0 replies, 7 likes


Statistics Papers: Generalizing Convolutional Neural Networks for Equivariance to Lie Groups on Arbitrary Continuous Data. http://arxiv.org/abs/2002.12880

0 replies, 7 likes


Kyle Cranmer: @jonkhler @_onionesque @erikjbekkers @MilesCranmer @wellingmax @DaniloJRezende There is also this work Generalizing Convolutional Neural Networks for Equivariance to Lie Groups on Arbitrary Continuous Data from Marc Finzi, Samuel Stanton, Pavel Izmailov,@andrewgwils https://arxiv.org/abs/2002.12880

0 replies, 2 likes


Miles Cranmer: 4/10 @erikjbekkers - https://arxiv.org/abs/1909.12057 (images/grids) & @m_finzi - https://arxiv.org/abs/2002.12880 (alt. technique + extends to point clouds + Hamiltonian GNs) - Enforce symmetry in CNNs over a variety of Lie groups (2D rotational symmetry ~ "SO(2) Lie group") https://t.co/zcLM1FdB4a

1 replies, 2 likes


Content

Found on Mar 02 2020 at https://arxiv.org/pdf/2002.12880.pdf

PDF content of a computer science paper: Generalizing Convolutional Neural Networks for Equivariance to Lie Groups on Arbitrary Continuous Data