Double-Scoring: Reliable Extraction of Strong Lottery Tickets
arXiv:2607.20555
2026
Architecture
1 ideas extracted · analyzed Aug 31, 2026
What the math gives to ML
The paper identifies a practical failure mode in strong lottery-ticket extraction: edge-popup must choose a layerwise sparsity level, and an incorrect fixed density can prevent the score optimizer from finding a useful mask. Its transferable contribution is an augmented score space in which each real edge is accompanied by a zero-weight dummy edge; fixed-density top-k selection in this enlarged space allows the number of active real edges to vary automatically. This converts a discrete layerwise sparsity search into ordinary score optimization while preserving the representational freedom of original-coordinate masks. The most direct ML test is to replace edge-popup's per-layer top-k mask with top-k selection over concatenated real and zero-weight candidate edges, keeping weights frozen and measuring accuracy versus density-hyperparameter sensitivity.
Ideas from this paper
△ Mechanism confirmed, baseline not beaten
2026
For each frozen weight tensor, append a second tensor of identically shaped zero weights and assign trainable scores to both the real and dummy edges. Select a fixed number of candidates by top-k score in the doubled space; real edges selected by the competition remain active, while selected dummy edges consume the quota without changing the network. The resulting number of active original edges is learned rather than imposed by a separate layerwise sparsity search.
Useful7/10
Difficulty4/10
Novelty7/10