Parameter-Level Attribution of Symmetry in Trained Networks Though Parameter-Wise Functional Sensitivity

arXiv:2608.24700v1 Announce Type: new Abstract: When a network has learned a function with a known symmetry, can that symmetry be moved through the parametrisation---is there a motion in parameter space realising the group action in function space? We formulate this as a lifting problem for the realisation map Phi:hetamapsto f_heta, and show that a smooth parameter-space action exists only if the tangent space to the function's symmetry orbit lies within the image of mathrm dPhi_heta, whose columns are the functional sensitivities of individual parameters. This condition is also sufficient for pointwise first-order lifting. Relaxing it in least squares yields two local parameter directions: one following the symmetry orbit, one descending towards the equivariant subspace, with residuals measuring what the parametrisation cannot reach. On a rotationally invariant classifier we find these directions induce their predicted function-space motion, but only locally: recomputed directions track the orbit and reduce the equivariance defect, while directions held fixed depart from both after training. The same holds for Hamiltonian neural networks trained on a rotationally symmetric potential, even though the architecture does not explicitly enforce the symmetry.

Related

• A Complete Symmetry Classification of Shallow ReLU Networks
• TPV: Parameter Perturbations Through the Lens of Test Prediction Variance
• Mechanistic Anomaly Detection via Functional Attribution
• Measuring Structured Predictability in Neural Training Dynamics: A Cross-Regime Study

Source: arXiv cs.LG | 2026-08-26