Action-Factored Multi-Agent Reinforcement Learning for Scalable Quantum Device Tuning
This work addresses the scalability bottleneck in tuning large quantum-dot arrays, a critical step toward building large-scale quantum processors.
The authors propose QADAPT, a multi-agent RL framework with a factored action space to decouple agents and minimize interference in quantum-dot array tuning. It achieves zero-shot generalization to unseen device sizes and constant convergence steps, enabling scalable calibration of large-scale quantum processors.
Cooperative multi-agent reinforcement learning is well suited to problems with large parameter spaces and exploitable local structure, such as the tuning of electrostatically-defined quantum-dot arrays. However, if parameter cross-talk is strong, a non-stationary environment from the perspective of any individual agent can destabilize learning - the same effect that plagues manual tuning of such systems. We propose using a factored representation of the action space, learned online, to decouple agents and minimize their interference. Our framework, QADAPT, uses this factorization to efficiently learn shared policies based on local measurements and rewards. With this modular strategy, we achieve zero-shot generalization to unseen quantum device sizes and maintain an approximately constant number of convergence steps to reach target regimes. This work provides a scalable route toward the rapid calibration of large-scale quantum processors.