Beyond the Birkhoff Polytope: Spectral-Sphere-Constrained Hyper-Connections
This addresses a specific bottleneck in neural network architecture design for researchers and practitioners, offering a more efficient and stable alternative to existing constrained methods.
The paper tackled the problem of unconstrained mixing in Hyper-Connections causing unstable training by proposing Spectral-Sphere-Constrained Hyper-Connections (sHC), which shifted the feasible set to a spectral norm sphere, allowing negative entries and eliminating unstable projections, resulting in improved expressivity and stability without identity degeneration.
Hyper-Connections (HC) generalize residual connections into multiple streams, employing residual matrices for cross-stream feature mixing to enrich model expressivity. However, unconstrained mixing disrupts the identity mapping property intrinsic to the residual connection, causing unstable training. To address this, Manifold-Constrained Hyper-Connections (mHC) and its variant restrict these matrices to the Birkhoff polytope (doubly stochastic matrices) via Sinkhorn iterations or permutation-based parameterizations. We reveal three limitations of this polytope constraint: (1) identity degeneration, where learned matrices collapse around the identity and diminish cross-stream interactions, (2) an expressivity bottleneck, as the non-negativity constraint prevents subtractive feature disentanglement, and (3) parameterization inefficiencies, manifesting as unstable Sinkhorn iterations or the factorial-scaling overhead of permutation-based parameterizations. To overcome these flaws, we propose Spectral-Sphere-Constrained Hyper-Connections (sHC). By geometrically shifting the feasible set from a rigid polytope to a spectral norm sphere, sHC allows negative entries, unlocking subtractive interactions for selective feature diversification. This shift eliminates unstable Sinkhorn projections and factorial parameterization, enabling expressive, non-degenerate residual matrices while preserving training stability.