ProTuner: Tuning Programs with Monte Carlo Tree Search
This addresses the challenge of program optimization for performance in specific domains like deep learning and image processing, representing an incremental improvement over existing methods.
The paper tackled the problem of tuning programs for high-performance deep learning and image processing by applying Monte Carlo Tree Search (MCTS), and the result showed that MCTS outperformed the state-of-the-art beam-search algorithm on 16 real benchmarks.
We explore applying the Monte Carlo Tree Search (MCTS) algorithm in a notoriously difficult task: tuning programs for high-performance deep learning and image processing. We build our framework on top of Halide and show that MCTS can outperform the state-of-the-art beam-search algorithm. Unlike beam search, which is guided by greedy intermediate performance comparisons between partial and less meaningful schedules, MCTS compares complete schedules and looks ahead before making any intermediate scheduling decision. We further explore modifications to the standard MCTS algorithm as well as combining real execution time measurements with the cost model. Our results show that MCTS can outperform beam search on a suite of 16 real benchmarks.