Degree-tail Preservation
Cross-source consensus on Degree-tail Preservation from 1 sources and 3 claims.
1 sources · 3 claims
How it works
Comparisons
Highlighted claims
- RNS preserves the power-law degree-distribution tail in sampled induced subgraphs with the same exponent and a scaled constant. — Implicit Regularization of Mini-Batch Training in Graph Neural Networks
- Samplers that distort the power-law exponent more also show higher batch gradient variance. — Implicit Regularization of Mini-Batch Training in Graph Neural Networks
- RNS is distinctive because it preserves uniform node statistics while discarding local graph structure inside each batch. — Implicit Regularization of Mini-Batch Training in Graph Neural Networks