Tag
This controlled study compares geometric algebra (Cl(3,0)) layers against a minimal scalarization baseline for SO(3)-equivariant vector learning, finding that geometric algebra adds no benefit for single-stage tasks but significantly beats scalarization in low-data regimes for deeply composed group operations.