Speaker
Description
Efficient stabilizer state preparation is essential in quantum algorithms, measurement-based quantum computation, and quantum error correction. We introduce graph decimation, a scalable formulation of stabilizer-state circuit synthesis as a sequential decision-making problem on graph states and a compact classical learning environment for quantum circuit synthesis. The method uses exact graph transformation rules to remove graph edges using $\mathrm{CX}$, $\mathrm{CY}$, and $\mathrm{CZ}$ gates, then reverses this sequence to obtain a preparation circuit. We compare greedy search, beam search, model-free reinforcement learning (RL), and an RL-guided Monte Carlo tree search (MCTS) method, QuSynth. Across logical code states from 9 to 144 physical qubits, the best found circuits match or improve published best circuits where available. For larger codes in this range, for which we are not aware of published results, the best circuits we find reduce two-qubit gate count by up to a factor of 2.5 relative to LC-optimized $\mathrm{CZ}$-only baselines while retaining low circuit depth.
| Topic | Machine learning techniques for experimental quantum information science |
|---|