Open Problems

Equivariance & Symmetry in Neural Networks

Benchmarking Equivariance Degradation Across Discretization, Domain Shift, and Hybrid Pipelines

Effect to explainPartly addressed
Possible candidate · 1/2 runs3 papers report this67% from 2025+

Generated automatically from the limitations stated in 3 papers (ICML, NeurIPS, ICLR), listed under Evidence. It is not a paper, and it does not come from papers submitted to CSPaper.

The problem

Exact equivariant architectures suffer numerical degradation when implemented on discretized grids, while learned approximate equivariance fails to extrapolate under test-time distribution shifts. Furthermore, inserting non-equivariant intermediate layers (such as high-dimensional feature GCNs) into geometric backbones breaks end-to-end SE(3) equivariance, risking pipeline instability. Because no systematic study evaluates these distinct methods and failure modes across shared out-of-distribution and discretized settings, practitioners have no principled way to know whether exact algebraic constraints, learned symmetries, or hybrid pipelines will retain their geometric guarantees in deployment.

Why it matters

Provides empirical bounds and selection criteria determining when exact geometric constraints are mandatory versus when learned or hybrid representations suffice under real-world discretization and domain shifts.

Ways to approach it

Prior-work checks are free with an account. Results someone already ran are shown to everyone.

  1. 1

    Benchmark exact group-equivariant models (e.g., steerable CNNs, Clifford/CSCNNs) against learned-equivariance baselines across varying grid discretizations and out-of-distribution datasets (e.g., medical imaging), measuring the empirical equivariance error metric and downstream task accuracy.

  2. 2

    Evaluate hybrid 3D point-cloud and graph pipelines that mix SE(3)-equivariant encoders with standard high-dimensional GCN feature extractors, measuring output variance and numerical stability under continuous rigid rotations.

  3. 3

    Quantify layer-wise symmetry preservation across multi-stage pipelines to isolate where equivariance breaks and measure whether lightweight projection layers restore end-to-end transformation stability.

Have a different approach?

Describe how you would tackle this problem and we'll look for papers that already do it. Free; your text stays private.

Free · 3 checks per day

Why it might fail

If downstream task accuracy on standard real-world tasks proves completely resilient to intermediate symmetry breaking and discretization errors, the practical value of measuring the equivariance gap would be diminished. Additionally, standardizing a unified equivariance error metric across disparate representation types (multivectors, irreps, and coordinate graphs) is technically challenging.

Evidence

Each paper's own statement of the limitation, verbatim.

Nearest existing work

Related open problems

Equivariance & Symmetry in Neural Networks

UnclassifiedOpen

Auditing the Gap Between Theoretical Equivariance Guarantees and Empirical Performance in Finite, Symmetry-Broken Architectures

Current theoretical guarantees for equivariant neural networks rely on assumptions that practical pipelines routinely violate: infinite network width, exact group representations, strict absence of symmetry breaking, and fully equivariant backbones. Practitioners regularly deploy approximate, learned, or symmetry-broken equivariant models alongside non-equivariant backbones (e.g., ResNet, Swin), where theoretical equivariance guarantees formally collapse. Because no unified evaluation has tested these isolated scope limitations across common benchmarks, it remains unknown whether reported empirical gains stem from geometric inductive biases or trivial capacity increases, and why learned equivariance fails to extrapolate under distribution shift.

Possible candidate · 2/4 runs5 papers report this40% from 2025+
Generated automatically, not curated by hand. Automated prior-work checks catch about a third of existing work, so treat this problem as a lead to investigate.