A General Framework for Equivariant Neural Networks on Reductive Lie Groups

Reductive Lie Groups, such as the orthogonal groups, the Lorentz group, or the unitary groups, play essential roles across scientific fields as diverse as high energy physics, quantum mechanics, quantum chromodynamics, molecular dynamics, computer vision, and imaging. In this paper, we present a general Equivariant Neural Network architecture capable of respecting the symmetries of the finite-dimensional representations of any reductive Lie Group G. Our approach generalizes the successful ACE and MACE architectures for atomistic point clouds to any data equivariant to a reductive Lie group action. We also introduce the lie-nn software library, which provides all the necessary tools to develop and implement such general G-equivariant neural networks. It implements routines for the reduction of generic tensor products of representations into irreducible representations, making it easy to apply our architecture to a wide range of problems and groups. The generality and performance of our approach are demonstrated by applying it to the tasks of top quark decay tagging (Lorentz group) and shape recognition (orthogonal group).

Paper

Similar papers

Peer review

Reviewer KJSX5/10 · confidence 5/52023-07-02

Summary

The paper presents a novel and highly general Equivariant Neural Network (ENN) architecture that is capable of respecting the symmetries of the finite-dimensional representations of any reductive Lie Group G. The proposed approach generalizes the successful ACE and MACE architectures for atomistic point clouds to any data equivariant to a reductive Lie group action. The authors demonstrate the generality and performance of their approach by applying it to two different tasks, namely top quark decay tagging (Lorentz group) and shape recognition (orthogonal group). The results presented in the paper are convincing and showcase the potential of the proposed architecture.

Strengths

* The paper is well-organized and well-written, providing a clear overview of the problem and proposed solution. * The paper addresses a significant and challenging problem by proposing a highly general ENN architecture that can handle symmetries of the finite-dimensional representations of any reductive Lie Group G, which contributes to the field of equivariant neural networks.

Weaknesses

Related concerns are discussed in the questions section.

Questions

* A more comprehensive discussion comparing methods based on Lie Groups, such as LieConv, should be included in the paper. * The results presented in the paper might not adequately convey the effectiveness of the proposed method. In order to reinforce the claims and emphasize the practical applicability of the approach, it would be valuable for the authors to incorporate comparisons of experimental results from methods like MACE and LieConv, using examples such as those from the QM9 dataset.

Rating

5: Borderline accept: Technically solid paper where reasons to accept outweigh reasons to reject, e.g., limited evaluation. Please use sparingly.

Confidence

5: You are absolutely certain about your assessment. You are very familiar with the related work and checked the math/other details carefully.

Soundness

3 good

Presentation

3 good

Contribution

3 good

Limitations

There are no potential negative societal impacts of this work.

Reviewer 5uTH6/10 · confidence 4/52023-07-06

Summary

The paper proposes a class of G-equivariant neural networks for reductive lie groups that generalizes the previous ACE and MCAE models (which are designed to be equivariant with respect to the orthogonal group O(3)) to more general irreducible Lie groups. A software library (Lie-NN) has also been developed and released for implementation of the proposed method. The experimental results on jet tagging and 3d point cloud classification support the authors' claims about the equivariant model improving model performance.

Strengths

- The proposed approach seems novel; I am not aware of other works that design a G-equivariant network in this way. - The paper is generally well-written and technically correct (although I have not performed an in-depth check for the proof of the universality theorem). Occasional variable definitions are missing but I had no trouble following the general flow of the paper. - The experimental results appear to validate the claimed advantages of the proposed method (although more discussion on the computational aspects is needed to assess practicality; see below).

Weaknesses

- Since this work is a generalized version of the previous ACE and MACE methods, a brief review of ACE and MACE would be helpful to understand and evaluate the novelty of the method. The current version is presented in a way that contains elements of previous existing methods, without a clear delineation of the new and original aspects of the proposed method. Exactly what is the precise nature of the extensions to ACE and MACE should be made clearer. - While the technical contents seem correct, some effort at providing intuitive explanation and justifications at key places would be helpful to understanding the paper. Descriptive figures, intuitive examples (e.g., basis including 1-particle basis or Clebsch-Gordan coefficients) come to mind as examples. In particular, I am curious about how to construct a 1-particle basis for specific Lie groups, e.g., SO(3). - The formulation in the case of the product group in Section 4 seems to be missing, although there is a mention of the product group in the introduction and in Section 5. Is it trivial to design an equivariant model for the product group G_1 x G_2 with the formulation used in this method? My initial impression is that there may be some subtleties involved, such as the order of the group actions. - A discussion of the computational aspects of this method is missing. When using this model, the Clebsch-Gordan coefficients must be calculated numerically (as mentioned in Section 5), and it would be helful to mention calculation times and errors. Does the calculation time vary depending on the number of basis components? I am also curious about how much time it takes for iteration on backpropagation compared to classic MLP models. - Continuing with the above comment, since the Clebsch-Gordan coefficients are numerically calculated, some numerical calculation errors are inevitable. Is the equivariance of the model maintained in the presence of numerical errors? If not, some discussion, even qualitative, should be provided (e.g., to what extent the model remains equivariant) although quantitative results (e.g., experimental results, figures) would clearly be preferable.

Questions

These have been mentioned for the most part in the weaknesses section.

Rating

6: Weak Accept: Technically solid, moderate-to-high impact paper, with no major concerns with respect to evaluation, resources, reproducibility, ethical considerations.

Confidence

4: You are confident in your assessment, but not absolutely certain. It is unlikely, but not impossible, that you did not understand some parts of the submission or that you are unfamiliar with some pieces of related work.

Soundness

3 good

Presentation

3 good

Contribution

3 good

Limitations

No potential negative societal impact of this work as far as I can tell. As mentioned earlier, computational aspects of the method would be helpful. I don't want the authors to feel compelled to show that this method is immediately computationally practical, as that is not the only measure of the value and worth of any new idea, but it would be helpful to indicate to other researchers on what the computational limitations are, and to possibly spur interest in finding improvements.

Reviewer Mk5u8/10 · confidence 4/52023-07-06

Summary

The authors generalize MACE, a point cloud network that uses higher-order interactions via tensor products of basis expansions of the features, to being equivariant to arbitrary reductive Lie groups. The paper shows that this setup inherits universality properties from MACE. A generic method to compute the a basis for the equivariant linear maps between tensor products of the representations is implemented. While the paper is mostly theoretical, the authors show strong performance on a SO(1, 3) equivariant task and a point cloud task.

Strengths

- The proposed method is an elegant generalization of a popular prior method - The paper is well-written and mostly easy to read. - The code is included - The proposed method could be a useful turn-key equivariance solution for researchers with a niche symmetry problem.

Weaknesses

- Figure 1 of the paper appears to suggest that the method works on E(3), but this can't be a reductive group, as the 4D homogenous representation is not decomposable as a sum of irreducibles (contradicting lines 67-68). I suppose the authors mean to write O(3) and treat the translation by canonicalization, but they should clarity that. - The point could experiment should include more baselines. - As the key contribution of the paper mostly lies in implementing the necessary computations for generic reductive Lie groups (sec 5), it would make sense to allocate more space to that and less to the prior sections, which are mostly already covered in prior works. Alternatively / in addition, it would be good to give more background on the material in section 5 in the appendix.

Questions

- It would be helpful to include/repeat the definition of B used in equation (20).

Rating

8: Strong Accept: Technically strong paper, with novel ideas, excellent impact on at least one area, or high-to-excellent impact on multiple areas, with excellent evaluation, resources, and reproducibility, and no unaddressed ethical considerations.

Confidence

4: You are confident in your assessment, but not absolutely certain. It is unlikely, but not impossible, that you did not understand some parts of the submission or that you are unfamiliar with some pieces of related work.

Soundness

4 excellent

Presentation

3 good

Contribution

3 good

Limitations

The authors should clarify better that not all Lie groups of interest are reductive.

Reviewer jt5w6/10 · confidence 2/52023-07-07

Summary

This paper proposes a framework for building equivariant neural networks on reductive lie groups. The proposed method first constructs a linear model for multi-set functions which is then symmetrised to generate a complete basis of equivariant multi-set functions. The model is also extended to a multi-layer architecture by a message-passing scheme.

Strengths

* The idea of generalizing ACE and MACE frameworks to arbitrary reductive Lie groups is new. * Proofs are given in the supplemental material to support theoretical claims in the paper. * A library is provided for developing G-equivariant neural networks.

Weaknesses

* The paper has some minor errors in writting. For instance, Table 2 caption should be on top of the table. Tables (e.g., Tabs. (2) and (3)) that report experimental results should be mentioned in the text. * Lack of comparison against state-of-the-art methods for validating the effectiveness of the proposed method on 3D shape recognition.

Questions

* The computations in Eqs. (7) together with (16) seem to be expensive. How these can be done in practice ? * It would be interesting to show the impact of higher order messages on the performance of G-MACE in terms of accuracy and computation time. I didn't find this study in the paper and supplemental material.

Rating

6: Weak Accept: Technically solid, moderate-to-high impact paper, with no major concerns with respect to evaluation, resources, reproducibility, ethical considerations.

Confidence

2: You are willing to defend your assessment, but it is quite likely that you did not understand the central parts of the submission or that you are unfamiliar with some pieces of related work. Math/other details were not carefully checked.

Soundness

3 good

Presentation

3 good

Contribution

3 good

Limitations

Limitations are discussed in the supplemental material.

Reviewer Mk5u2023-08-11

Thanks for the response

I thank the authors for their response. My score remains unchanged.

Reviewer Mk5u2023-08-11

Appropriate to refer to arxiv?

Dear authors and AC, Given that the reviewers are not allowed to seek out this paper on arxiv and this conference does not permit uploading a revised version of the paper, I believe it is inappropriate to refer to arxiv, or to any existing updated version of the paper in general. Perhaps the authors can remove the reference to arxiv in their rebuttal and rephrase their rebuttals from "we have changed/fixed" to "we will change/fix". Thanks

Authorsrebuttal2023-08-11

Dear Reviewers and AC, We now realize that this was indeed inappropriate and put this mistake on our haste and confusion on the last minutes before the deadline. For future readers, please ignore this line in the rebuttal. If the AC could allow us to remove that line from the rebuttal as we can not do it now?

Area Chair 3D2Y2023-08-15

ArXiv

I agree that it would have been better not to refer to the arxiv preprint, since reviewers are expected not to look at the de-anonimized arxiv paper. I cannot remove the line in the review, but I expect that the reviewers will be able to disregard it.

Reviewer KJSX2023-08-15

Thanks for your response

I appreciate the additional details provided in the rebuttal, as they have addressed the majority of my questions and concerns. Therefore, I will increase the rating.

Reviewer 5uTH2023-08-15

I appreciate the follow-up to my queries; my questions and concerns have been sufficiently addressed. I'm still left with the impression that computational and numerical considerations for this method are important and yet are not mentioned as prominently as they should be in the main body of the paper (only in the appendix). Perhaps there's not much the authors can do about this, since revisions to the main body of the paper are not allowed at this stage.

Authorsrebuttal2023-08-15

We are glad to know that our response was satisfactory to you. We plan to include all the new experiments related to computational and numerical considerations in the main part of the paper. This will be done as soon as we are permitted an additional page for the final version. We hope this plan is satisfactory to you.

Reviewer jt5w2023-08-21

I thank the authors for their adequate answers. I maintain my original rating.

Program Chairsdecision2023-09-21

Decision

Accept (poster)

© 2026 NYSGPT2525 LLC