Learning to Rank Tensor Network Contraction Plans for GPU-Accelerated Quantum Circuit Simulation

The authors propose a method that learns to predict the relative performance of contraction plans on GPUs, where plans with similar theoretical cost can differ significantly due to factors like parallelism and memory traffic. The approach likely uses a ranking model trained on benchmark data from quantum circuit simulations. This is important because choosing a better contraction plan can substantially speed up classical simulation of quantum circuits, which is essential for validating near-term quantum algorithms. The work targets the practical bottleneck of GPU execution, moving beyond traditional cost models that only consider flop counts.