Reinforced Genetic Algorithm Learning for Optimizing Computation Graphs
This work addresses the challenge of efficiently optimizing computation graphs in compilers for machine learning practitioners, offering a scalable solution that reduces optimization time from hours to seconds.
The paper tackles the problem of optimizing neural network computation graphs for execution cost by introducing a deep reinforcement learning approach that trains offline and generalizes to unseen graphs, achieving significant improvements in running time and peak memory usage compared to classical and other learning-based methods.
We present a deep reinforcement learning approach to minimizing the execution cost of neural network computation graphs in an optimizing compiler. Unlike earlier learning-based works that require training the optimizer on the same graph to be optimized, we propose a learning approach that trains an optimizer offline and then generalizes to previously unseen graphs without further training. This allows our approach to produce high-quality execution decisions on real-world TensorFlow graphs in seconds instead of hours. We consider two optimization tasks for computation graphs: minimizing running time and peak memory usage. In comparison to an extensive set of baselines, our approach achieves significant improvements over classical and other learning-based methods on these two tasks.