model-equivariance-auditor

You can test neural network equivariance by applying specified symmetry group transforms to inputs and measuring output deviations. This skill automates end-to-end equivariance testing across common symmetry groups like SO3, SE3, and permutation, generating structured diagnostic reports.

To pinpoint broken equivariance, use layer-wise probing to measure symmetry deviations at each layer. This skill provides layer-wise analysis to identify exactly where equivariance is broken in your neural network and offers actionable recommendations to guide fixes.

Yes, you can audit gradient equivariance to verify that training dynamics preserve symmetry. This skill performs gradient-equivariance checks to ensure that gradients respect the intended symmetry groups during model training, producing automated diagnostic reports.

The best way to generate a diagnostic report for model symmetry violations is to run an automated equivariance audit. This skill computes absolute and relative error metrics across symmetry transforms and produces a structured audit report with actionable debugging recommendations.

Equivariance testing works for both geometric rotations and permutation symmetries. This skill supports auditing neural networks with respect to common symmetry groups including SO3, SE3, and permutation transformations, covering both geometric and discrete symmetries.

To run an equivariance audit, you need a test harness capable of applying input and output transforms and computing error metrics. This skill requires a test harness setup that can apply symmetry group transforms and collect results for automated report generation.