A Wigner-Eckart Theorem for Group Equivariant Convolution Kernels

Group equivariant convolutional networks (GCNNs) endow classical convolutional networks with additional symmetry priors, which can lead to a considerably improved performance. Recent advances in the theoretical description of GCNNs revealed that such models can generally be understood as performing convolutions with G-steerable kernels, that is, kernels that satisfy an equivariance constraint themselves. While the G-steerability constraint has been derived, it has to date only been solved for specific use cases - a general characterization of G-steerable kernel spaces is still missing. This work provides such a characterization for the practically relevant case of G being any compact group. Our investigation is motivated by a striking analogy between the constraints underlying steerable kernels on the one hand and spherical tensor operators from quantum mechanics on the other hand. By generalizing the famous Wigner-Eckart theorem for spherical tensor operators, we prove that steerable kernel spaces are fully understood and parameterized in terms of 1) generalized reduced matrix elements, 2) Clebsch-Gordan coefficients, and 3) harmonic basis functions on homogeneous spaces.

[1]  Max Welling,et al.  Spherical CNNs , 2018, ICLR.

[2]  Li Li,et al.  Tensor Field Networks: Rotation- and Translation-Equivariant Neural Networks for 3D Point Clouds , 2018, ArXiv.

[3]  Maurice Weiler,et al.  General E(2)-Equivariant Steerable CNNs , 2019, NeurIPS.

[4]  V. K. Agrawala.  Wigner–Eckart theorem for an arbitrary group or Lie algebra , 1980 .

[5]  H. Reiter.  Classical Harmonic Analysis and Locally Compact Groups , 1968 .

[6]  Zhen Lin,et al.  Clebsch-Gordan Nets: a Fully Fourier Space Spherical Convolutional Neural Network , 2018, NeurIPS.

[7]  Wigner-Eckart theorem and Jordan-Schwinger representation for infinite-dimensional representations of the Lorentz group , 2015, 1509.05633.

[8]  John B. Conway.  A Course in Point Set Topology , 2016 .

[9]  Charles R. Johnson,et al.  Matrix analysis , 1985, Statistical Inference for Engineers and Data Scientists.

[10]  J. Herod.  Introduction to Hilbert spaces with applications , 1990 .

[11]  Hans Burkhardt,et al.  Learning Equivariant Functions with Matrix Valued Kernels , 2007, J. Mach. Learn. Res..

[12]  Daniel E. Worrall,et al.  Deep Scale-spaces: Equivariance Over Scale , 2019, NeurIPS.

[13]  Yacov Hel-Or,et al.  Canonical Decomposition of Steerable Functions , 2004, Journal of Mathematical Imaging and Vision.

[14]  Elmar Eisemann,et al.  CNNs on surfaces using rotation-equivariant features , 2020, ACM Trans. Graph..

[15]  David W. Miller,et al.  Lorentz Group Equivariant Neural Network for Particle Physics , 2020, ICML.

[16]  J. Gallier,et al.  Differential Geometry and Lie Groups , 2020, Geometry and Computing.

[17]  Casimir Wierzynski,et al.  Learning Irreducible Representations of Noncommutative Lie Groups , 2020, ArXiv.

[18]  Erik J Bekkers.  B-Spline CNNs on Lie Groups , 2020, ICLR.

[19]  Ivan Sosnovik,et al.  Scale-Equivariant Steerable Networks , 2020, ICLR.

[20]  N. Bourbaki.  General Topology: Chapters 1–4 , 1989 .

[21]  Max Welling,et al.  Group Equivariant Convolutional Networks , 2016, ICML.

[22]  Max Welling,et al.  Gauge Equivariant Mesh CNNs: Anisotropic convolutions on geometric graphs , 2020, ICLR.

[23]  Stephan J. Garbin,et al.  Harmonic Networks: Deep Translation and Rotation Equivariance , 2016, 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR).

[24]  Edward H. Adelson,et al.  The Design and Use of Steerable Filters , 1991, IEEE Trans. Pattern Anal. Mach. Intell..

[25]  Risi Kondor,et al.  On the Generalization of Equivariance and Convolution in Neural Networks to the Action of Compact Groups , 2018, ICML.

[26]  E. Guth,et al.  Gruppentheorie und ihre Anwendung auf die Quantenmechanik der Atomspektren , 1932 .

[27]  Wilson A. Sutherland,et al.  Introduction to Metric and Topological Spaces , 1975 .

[28]  I. Segal.  AN EXTENSION OF PLANCHEREL'S FORMULA TO SEPARABLE UNIMODULAR GROUPS , 1950 .

[29]  Max Welling,et al.  Steerable CNNs , 2016, ICLR.

[30]  Maurice Weiler,et al.  A General Theory of Equivariant CNNs on Homogeneous Spaces , 2018, NeurIPS.

[31]  Maurice Weiler,et al.  Learning Steerable Filters for Rotation Equivariant CNNs , 2017, 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition.

[32]  Risi Kondor,et al.  N-body Networks: a Covariant Hierarchical Neural Network Architecture for Learning Atomic Potentials , 2018, ArXiv.

[33]  David B. A. Epstein,et al.  Dense Steerable Filter CNNs for Exploiting Rotational Symmetry in Histology Images , 2020, IEEE Transactions on Medical Imaging.

[34]  Max Welling,et al.  Gauge Equivariant Convolutional Networks and the Icosahedral CNN 1 , 2019 .

[35]  Max Welling,et al.  3D Steerable CNNs: Learning Rotationally Equivariant Features in Volumetric Data , 2018, NeurIPS.

[36]  F. I. Mautner.  Note on the Fourier inversion formula on groups , 1955 .

[37]  T. MacRobert.  Spherical harmonics : an elementary treatise on harmonic functions , 1927 .

[38]  Risi Kondor,et al.  Cormorant: Covariant Molecular Neural Networks , 2019, NeurIPS.