AI and the World's Hard Problems · Quantum physics & quantum computing · arXiv · 2026-10-19
ConQUESTA: Topology-Aware Post-Training Structural Pruning of Quantum Neural Networks
Deploying quantum neural networks on sparse hardware requires cutting two-qubit gates, SWAPs, and depth without losing accuracy. The authors propose ConQUESTA, a post-training pruning method that scores entanglers by task saliency and compiler cost, then searches whole pruning masks under an accuracy constraint, transpiling each mask across several coupling topologies. At 55% pruning, mean resource reductions were 61.6% on Breast Cancer, 72.0% on Digits01, and 82.2% on Iris, and fixed-mask recovery raised feasible runs from 16 to 22 of 24.