A graph policy optimization method combines multi-hop GNN embeddings for global fill-in with local symbolic factorization feedback and an adaptive saturation function, achieving 29.3 mean fill-in and 31.3 peak memory reductions over baselines on SuiteSparse matrices.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Learning Fill-in Reduction Ordering via Graph Policy Optimization for Sparse Matrices
A graph policy optimization method combines multi-hop GNN embeddings for global fill-in with local symbolic factorization feedback and an adaptive saturation function, achieving 29.3 mean fill-in and 31.3 peak memory reductions over baselines on SuiteSparse matrices.