Replay-Buffer Engineering for Quantum Circuit Optimization

Paper: 2604.21863 Authors: Research Team Categories: cs.AI, cs.QU

Problem

Quantum circuit compilation faces unique challenges:

• Exponential search space for gate ordering
• Hardware-specific constraints (connectivity, calibration)
• Trade-offs between circuit depth and gate count
• Limited quantum hardware access for training

Replay-Buffer Engineering

Adapting classical RL techniques for quantum optimization:

Key Innovation

Apply prioritized experience replay to quantum circuit synthesis:

1. Quantum State Encoding

   • Represent circuit as sequence of operations
   • Encode circuit properties (depth, fidelity, gate count)
   • Track quantum resource usage

2. Prioritized Sampling

   • Priority based on circuit improvement potential
   • Balance exploration vs. exploitation
   • Diverse sampling for better generalization

3. Buffer Management

   • Novelty-based prioritization
   • Diversity maintenance in buffer
   • Efficient storage for circuit representations

Results

Metric	Standard RL	Replay-Buffer RL
Circuit Depth	124 gates	87 gates
Training Samples	50K	12K
Convergence Speed	8 hours	2 hours
Fidelity	92%	97%

Takeaways

• RL techniques transfer well from classical to quantum domains
• Prioritized replay accelerates quantum circuit learning
• Sample efficiency is crucial given quantum hardware scarcity
• Buffer engineering matters for quantum RL