Gate Fusion Accelerates Quantum Circuit Simulation Throughput

Yoshiaki Kawase (arXiv v1, Mar 3, 2026) presents a gate-fusion method to speed classical simulation of quantum machine learning. Fusing consecutive gates in forward and backward passes yields ≈20× throughput for 12+ qubits and >30× on a mid-range GPU; combined with gradient checkpointing it enables training a 20-qubit, 1,000-layer model (60,000 parameters) with 1,000 samples in about 20 minutes.