Precision autotuning for linear solvers via contextual bandit-based RL

Erin Carson, Xinye Chen

We propose a reinforcement learning (RL) framework for adaptive precision tuning for linear solvers, which can be extended to general algorithms. The framework is formulated as a c…