OpenNovelty.org

Achieving Approximate Symmetry Is Exponentially Easier than Exact Symmetry

ICLR 2026 Conference SubmissionAnonymous Authors
OpenReview Score: 5.6Download Report PDF
symmetryinvariancerelaxed equivariancecomplexitytheory
Abstract:

Enforcing exact symmetry in machine learning models often yields significant gains in scientific applications, serving as a powerful inductive bias. However, recent work suggests that relying on approximate symmetry can offer greater flexibility and robustness. Despite promising empirical evidence, there has been little theoretical understanding, and in particular, a direct comparison between exact and approximate symmetry is missing from the literature. In this paper, we initiate this study by asking: What is the cost of enforcing exact versus approximate symmetry? To address this question, we introduce averaging complexity, a framework for quantifying the cost of enforcing symmetry via averaging. Our main result is an exponential separation: under standard conditions, achieving exact symmetry requires linear averaging complexity, whereas approximate symmetry can be attained with only logarithmic averaging complexity. To the best of our knowledge, this provides the first theoretical separation of these two cases, formally justifying why approximate symmetry may be preferable in practice. Beyond this, our tools and techniques may be of independent interest for the broader study of symmetries in machine learning.

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 an averaging complexity framework to quantify the cost of enforcing exact versus approximate symmetry, establishing an exponential separation: linear complexity for exact symmetry versus logarithmic for approximate. Within the taxonomy, it occupies the 'Averaging Complexity and Exact versus Approximate Symmetry' leaf under Theoretical Foundations, where it is currently the sole paper. This positioning reflects a sparse research direction focused specifically on complexity-theoretic distinctions between symmetry enforcement regimes, rather than the more crowded applied methods branches.

The taxonomy reveals that neighboring work concentrates on applied frame averaging techniques (e.g., Minimal Frame Averaging, Faenet Frame Averaging) and equivariance theory establishing inductive bias benefits. The paper's theoretical lens on averaging cost contrasts with these implementation-focused directions. The Equivariance and Invariance Theory sibling leaf addresses why symmetry helps generalization but excludes complexity analysis, while Frame Averaging Techniques develop practical algorithms without formal cost bounds. This work thus fills a gap between foundational equivariance justifications and algorithmic implementations by quantifying the computational trade-offs inherent in averaging-based symmetry enforcement.

Among 21 candidates examined across three contributions, no refutable prior work was identified. The exponential separation result examined 1 candidate with 0 refutations; the averaging complexity framework examined 10 candidates with 0 refutations; explicit bounds examined 10 candidates with 0 refutations. This suggests that within the limited search scope—focused on top-K semantic matches and citation expansion—no prior work directly establishes comparable complexity separations or averaging cost frameworks. The absence of refutations across all contributions indicates the theoretical angle appears relatively unexplored in the examined literature.

Given the limited search scope of 21 candidates, the analysis captures semantic neighbors and cited works but cannot claim exhaustive coverage of all theoretical complexity studies in symmetry enforcement. The taxonomy structure and contribution-level statistics together suggest the paper addresses a theoretically underexplored question, though broader searches in computational complexity or approximation theory might reveal additional context not captured here.

Taxonomy

Core-task Taxonomy Papers
26
3
Claimed Contributions
21
Contribution Candidate Papers Compared
0
Refutable Paper

Research Landscape Overview

Core task: enforcing symmetry in machine learning models via averaging. The field organizes around three main branches that reflect distinct perspectives on how symmetry constraints can be integrated into learning systems. Theoretical Foundations of Symmetry Enforcement examines the mathematical underpinnings, including questions of averaging complexity and the distinction between exact versus approximate symmetry—where works like Approximate Symmetry[0] and Symmetry Generalisation ML[1] explore how models handle imperfect or partial symmetries. Applied Methods for Symmetry Enforcement focuses on algorithmic techniques such as frame averaging (e.g., Frame Averaging Networks[9], Faenet Frame Averaging[6], Minimal Frame Averaging[7]) and model averaging strategies (Model Averaging DML[4]), which provide practical mechanisms to enforce invariance or equivariance properties. Domain Applications of Symmetry Enforcement spans diverse areas from molecular modeling (Symmetry Atom Mapping[5], Multiscale Molecular Modeling[12]) to physics-inspired problems (Average SPT Phases[3], Quantum Ising Symmetry[13]) and even pose estimation (ES6D Pose Regression[14]), demonstrating how symmetry constraints improve generalization and physical consistency across contexts. A particularly active line of work centers on frame averaging methods, which enforce symmetry by averaging model outputs over group transformations. This approach trades computational cost for guaranteed equivariance, with recent efforts like Minimal Frame Averaging[7] seeking to reduce the averaging burden while maintaining theoretical guarantees. In contrast, Approximate Symmetry[0] sits within the theoretical branch examining scenarios where exact symmetry is computationally prohibitive or the underlying data exhibits only approximate invariance. This work connects closely to Symmetry Generalisation ML[1], which investigates how models learn and generalize symmetry patterns, and to Average Symmetry Topological[2], which explores symmetry in topological contexts. By addressing the complexity-accuracy trade-off inherent in averaging-based symmetry enforcement, Approximate Symmetry[0] bridges theoretical insights about relaxed symmetry conditions with the practical constraints faced by applied methods, offering a complementary perspective to exact frame averaging techniques.

Claimed Contributions

Exponential separation between exact and approximate symmetry enforcement

The authors prove that enforcing exact symmetry in machine learning models requires averaging complexity linear in the group size, while approximate symmetry requires only logarithmic complexity. This exponential gap provides the first theoretical separation between these two widely used approaches, formally justifying why approximate symmetry may be preferable in practice.

1 retrieved paper
Averaging complexity framework for quantifying symmetry enforcement cost

The authors introduce a formal framework called averaging complexity that measures the cost of enforcing symmetry through the number of action queries needed in an averaging scheme. This framework enables rigorous comparison between exact and approximate symmetry enforcement and may be of independent interest for broader studies in geometric machine learning.

10 retrieved papers
Explicit bounds on averaging complexity for exact and approximate symmetry

The authors establish that exact symmetry enforcement requires linear averaging complexity (equal to the group size under certain conditions), while weak and strong approximate symmetry enforcement both achieve O(log|G|/ε) averaging complexity. These bounds are shown to be tight up to constants.

10 retrieved papers

Core Task Comparisons

Comparisons with papers in the same taxonomy category

Within the taxonomy built over the current TopK core-task papers, the original paper is assigned to a leaf with no direct siblings and no cousin branches under the same grandparent topic. In this retrieved landscape, it appears structurally isolated, which is one partial signal of novelty, but still constrained by search coverage and taxonomy granularity.

Contribution Analysis

Detailed comparisons for each claimed contribution

Contribution

Exponential separation between exact and approximate symmetry enforcement

The authors prove that enforcing exact symmetry in machine learning models requires averaging complexity linear in the group size, while approximate symmetry requires only logarithmic complexity. This exponential gap provides the first theoretical separation between these two widely used approaches, formally justifying why approximate symmetry may be preferable in practice.

Contribution

Averaging complexity framework for quantifying symmetry enforcement cost

The authors introduce a formal framework called averaging complexity that measures the cost of enforcing symmetry through the number of action queries needed in an averaging scheme. This framework enables rigorous comparison between exact and approximate symmetry enforcement and may be of independent interest for broader studies in geometric machine learning.

Contribution

Explicit bounds on averaging complexity for exact and approximate symmetry

The authors establish that exact symmetry enforcement requires linear averaging complexity (equal to the group size under certain conditions), while weak and strong approximate symmetry enforcement both achieve O(log|G|/ε) averaging complexity. These bounds are shown to be tight up to constants.

Table of Contents