A reinforcement learning search discovers hybrid, hardware- and distribution-tuned iterative algorithms for matrix functions, with a random-matrix-theory generalization guarantee and experiments showing speedups over baselines.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
MatRL: Provably Generalizable Iterative Algorithm Discovery via Monte-Carlo Tree Search
A reinforcement learning search discovers hybrid, hardware- and distribution-tuned iterative algorithms for matrix functions, with a random-matrix-theory generalization guarantee and experiments showing speedups over baselines.