OpenNovelty.org

Any-Subgroup Equivariant Networks via Symmetry Breaking

ICLR 2026 Conference SubmissionAnonymous Authors
OpenReview Score: 6.5Download Report PDF
equivariancesymmetry breakinggraph neural networkssymmetry
Abstract:

The inclusion of symmetries as an inductive bias, known as equivariance, often improves generalization on geometric data (e.g. grids, sets, and graphs). However, equivariant architectures are usually highly constrained, designed for symmetries chosen a priori, and not applicable to datasets with other symmetries. This precludes the development of flexible, multi-modal foundation models capable of processing diverse data equivariantly. In this work, we build a single model --- the Any-Subgroup Equivariant Network (ASEN) --- that can be simultaneously equivariant to several groups, simply by modulating a certain auxiliary input feature. In particular, we start with a fully permutation-equivariant base model, and then obtain subgroup equivariance by using a symmetry-breaking input whose automorphism group is that subgroup. However, finding an input with the desired automorphism group is computationally hard. We overcome this by relaxing from exact to approximate symmetry breaking, leveraging the notion of 2-closure to derive fast algorithms. Theoretically, we show that our subgroup-equivariant networks can simulate equivariant MLPs, and their universality can be guaranteed if the base model is universal. Empirically, we validate our method on symmetry selection for graph and image tasks, as well as multitask and transfer learning for sequence tasks, showing that a single network equivariant to multiple permutation subgroups outperforms both separate equivariant models and a single non-equivariant model.

Disclaimer
This report is AI-GENERATED using Large Language Models and WisPaper (A scholar search engine). It analyzes academic papers' tasks and contributions against retrieved prior work. While this system identifies POTENTIAL overlaps and novel directions, ITS COVERAGE IS NOT EXHAUSTIVE AND JUDGMENTS ARE APPROXIMATE. These results are intended to assist human reviewers and SHOULD NOT be relied upon as a definitive verdict on novelty.
NOTE that some papers exist in multiple, slightly different versions (e.g., with different titles or URLs). The system may retrieve several versions of the same underlying work. The current automated pipeline does not reliably align or distinguish these cases, so human reviewers will need to disambiguate them manually.
If you have any questions, please contact: mingzhang23@m.fudan.edu.cn

Overview

Overall Novelty Assessment

The paper introduces Any-Subgroup Equivariant Networks (ASEN), a framework enabling a single model to achieve equivariance to multiple permutation subgroups by modulating auxiliary inputs. It resides in the 'Multi-Subgroup and Flexible Equivariance Mechanisms' leaf, which contains four papers total (including this one). This leaf sits within 'Architecture Design and Construction Methods', a moderately populated branch addressing practical network-building strategies. The small leaf size suggests this specific direction—simultaneous multi-subgroup equivariance via symmetry-breaking inputs—is relatively sparse compared to broader equivariance research, though the parent branch reflects sustained interest in flexible architectural solutions.

The taxonomy reveals neighboring leaves focused on permutation-equivariant layer constructions (five papers), non-linear extensions (two papers), and approximate equivariance (one paper). These adjacent directions tackle complementary challenges: fixed-group layer design, attention mechanisms for homogeneous spaces, and relaxed symmetry constraints. The original paper bridges these areas by starting from full permutation equivariance (a common baseline in layer constructions) and then achieving subgroup equivariance through approximate symmetry breaking. This positions ASEN at the intersection of flexible multi-group mechanisms and approximate methods, connecting modular construction principles (seen in sibling papers) with computational relaxation strategies.

Among fifteen candidates examined, the ASEN framework contribution (four candidates, zero refutations) and theoretical guarantees (ten candidates, zero refutations) appear relatively novel within this limited search scope. However, the approximate symmetry breaking via 2-closure contribution shows one refutable candidate among one examined, indicating prior work addresses similar computational relaxation techniques. The small candidate pool (fifteen total) means these statistics reflect top-K semantic matches and immediate citations, not exhaustive coverage. The framework's novelty seems strongest in its unified multi-subgroup approach, while the 2-closure algorithmic component overlaps more substantially with existing approximate methods.

Based on this limited analysis of fifteen candidates across three contributions, the work appears to occupy a relatively sparse research direction within the broader equivariance landscape. The multi-subgroup flexibility represents a less-explored architectural strategy compared to single-group designs, though the computational techniques for achieving it (approximate symmetry breaking) connect to established approximation literature. The assessment is constrained by the search scope and does not capture potential overlaps outside the top semantic matches or citation network examined.

Taxonomy

Core-task Taxonomy Papers
50
3
Claimed Contributions
15
Contribution Candidate Papers Compared
1
Refutable Paper

Research Landscape Overview

Core task: building flexible equivariant networks for multiple permutation subgroups. The field has organized itself around several complementary directions. Theoretical Foundations of Equivariance explores the mathematical underpinnings—group representations, invariant theory, and algebraic structures—that justify why and how symmetries can be encoded in neural architectures. Architecture Design and Construction Methods focuses on practical network-building strategies, including how to compose layers that respect one or many subgroup symmetries simultaneously, as seen in works like Modular PE-Structured[43] and Probabilistic Symmetrization[16]. Domain-Specific Applications and Implementations demonstrate these ideas in areas such as quantum systems (Quantum Graph Neural[14]), particle physics (Matrix-Element Likelihoods[9]), and combinatorial problems, while Computational Methods and Algorithmic Techniques address efficiency and scalability. Related Mathematical and Cryptographic Structures and Survey and Review Literature round out the taxonomy by connecting equivariance to broader mathematical contexts and summarizing progress across subfields. A particularly active line of work centers on multi-subgroup and flexible equivariance mechanisms, where the challenge is to design architectures that can adapt to or simultaneously respect several distinct permutation symmetries rather than committing to a single fixed group. Any-Subgroup Equivariant[0] exemplifies this direction by proposing methods that handle arbitrary subgroups within a unified framework, contrasting with earlier approaches that hardwired a single symmetry. Nearby efforts such as Modular PE-Structured[43] emphasize modular construction principles to combine different equivariance constraints, while Symmetry Invariant Design[46] explores how to encode invariance properties directly into layer design. These works collectively address the trade-off between expressiveness and computational cost: richer symmetry handling can improve generalization (Compositional Generalization[2]) but may require more sophisticated parameterizations or approximations (Approximately Equivariant[11]). The original paper sits squarely in this flexible multi-subgroup cluster, offering techniques that extend beyond fixed-group methods and align closely with modular and probabilistic symmetrization strategies.

Claimed Contributions

Any-Subgroup Equivariant Networks (ASEN) framework

The authors propose ASEN, a framework that constructs a single flexible model capable of being equivariant to multiple different symmetry groups by using a symmetry-breaking auxiliary input whose automorphism group matches the desired subgroup. This overcomes the inflexibility of traditional equivariant architectures that are designed for one specific symmetry group.

4 retrieved papers
Approximate symmetry breaking via 2-closure

The authors develop a practical algorithm using the 2-closure notion to construct symmetry-breaking inputs (positional and edge features) with approximately the desired automorphism group, making the framework computationally tractable when exact symmetry breaking is hard.

1 retrieved paper
Can Refute
Theoretical guarantees on expressivity and universality

The authors prove that ASEN can approximate equivariant MLPs to arbitrary accuracy and inherits universality properties from its base model, establishing formal expressivity guarantees for the framework.

10 retrieved papers

Core Task Comparisons

Comparisons with papers in the same taxonomy category

Contribution Analysis

Detailed comparisons for each claimed contribution

Contribution

Any-Subgroup Equivariant Networks (ASEN) framework

The authors propose ASEN, a framework that constructs a single flexible model capable of being equivariant to multiple different symmetry groups by using a symmetry-breaking auxiliary input whose automorphism group matches the desired subgroup. This overcomes the inflexibility of traditional equivariant architectures that are designed for one specific symmetry group.

Contribution

Approximate symmetry breaking via 2-closure

The authors develop a practical algorithm using the 2-closure notion to construct symmetry-breaking inputs (positional and edge features) with approximately the desired automorphism group, making the framework computationally tractable when exact symmetry breaking is hard.

Contribution

Theoretical guarantees on expressivity and universality

The authors prove that ASEN can approximate equivariant MLPs to arbitrary accuracy and inherits universality properties from its base model, establishing formal expressivity guarantees for the framework.

Table of Contents