| name | quantum-state-preparation-nn |
| category | quantum-computing |
| description | Neural network-based quantum state preparation methodology from arXiv:2605.31006. Trains classical neural networks to map input data directly to quantum circuit parameters, avoiding per-instance variational optimization. Achieves 0.992 fidelity on unseen images with 5000x runtime reduction. |
| source | arXiv:2605.31006 |
| source_title | Quantum State Preparation via Neural Network Encoding in Quantum Machine Learning |
| source_author | Kevin W. Aoun et al. |
| keywords | ["quantum state preparation","neural network encoding","quantum machine learning","amplitude encoding","variational circuits"] |
Neural Network Quantum State Preparation
Overview
Methodology for scalable quantum state preparation that replaces per-data-instance variational optimization with a single trained neural network mapping.
Trigger: When facing quantum state preparation bottlenecks, designing QML data loading pipelines, or optimizing amplitude encoding workflows.
arXiv: 2605.31006 | Author: Kevin W. Aoun et al.
Core Method
Problem
Amplitude encoding can represent 2ⁿ-dimensional data using n qubits, but preparing arbitrary states requires variational optimization of parameterized quantum circuits for each data instance — prohibitively expensive at scale.
Solution
Train a classical neural network to map input data directly to the continuous parameters of a fixed quantum circuit:
- Offline training: Optimize neural network weights on training dataset
- Single inference: Encode new inputs via one forward pass through the network
- Fixed circuit: Apply predicted parameters to predetermined quantum circuit structure
Performance
- Fidelity up to 0.992 on unseen MNIST/Fashion-MNIST images
- Per-data-instance runtime reduced by 5000x+
- All optimization performed once during training phase
Implementation Steps
- Design fixed parameterized quantum circuit (ansatz)
- Train classical neural network: input → circuit parameters
- Validate on held-out test set for generalization
- Deploy: new data → NN inference → quantum circuit execution
Pitfalls
- Fixed ansatz may limit expressivity for complex data distributions
- Neural network capacity must match circuit parameter count
- Generalization bounds depend on training data coverage
- Circuit depth constraints limit achievable fidelity
Verification Steps
- Measure fidelity on held-out test set
- Verify runtime scaling matches O(1) per instance after training
- Check generalization across data distribution shifts
- Benchmark against variational baseline for quality comparison
