WRSWorld Research StateContribute
WORLD RESEARCH STATE

Xie et al.: doubly-stochastic constrained hyper-connections

Contribution grade

Neural-network residual architecture · Human-designed research method or resource

Reference score95,000Provisional editorial estimate of this exact contribution; its grade is assessed separately.

What changed

Constrained multi-stream residual mixing through Sinkhorn-normalized nonnegative matrices and integrated fused kernels, recomputation and communication overlap to make expanded residual connections more stable at scale.

Scientific scope

Constrained multi-stream residual mixing through Sinkhorn-normalized nonnegative matrices and integrated fused kernels, recomputation and communication overlap to make expanded residual connections more stable at scale. Hyper-Connections and Sinkhorn-Knopp are inherited. Doubly stochastic mixing is not an exact identity map or a proof of full nonlinear network stability. Twenty practical normalization iterations only approximate the limiting constraint; norm contraction does not guarantee avoidance of collapse.

Evidence

The linked original sources describe the stated contribution and its evidence limits. WRS has not independently rerun the experiments or certified a proof. Hyper-Connections and Sinkhorn-Knopp are inherited. Doubly stochastic mixing is not an exact identity map or a proof of full nonlinear network stability. Twenty practical normalization iterations only approximate the limiting constraint; norm contraction does not guarantee avoidance of collapse.

Contributors

Provisional WRS assessment, edition 2026-10-16.1. Independent expert calibration is pending. This grade does not certify correctness or assign ownership.

Full scientific record · Browse contribution grades · Request a correction