| name | hybrid-quantum-neural-phase-recognition |
| description | Hybrid quantum-classical neural network for quantum phase recognition - jointly trains shallow parameterized quantum circuit with classical neural network, reduces sample complexity by ~10x, distinguishes topological phases on superconducting hardware |
| tags | ["quantum","neural-network","phase-recognition","topological-phase","hybrid-quantum-classical","surface-code","sample-efficient","superconducting-hardware","measurement-basis","trainable-transformation"] |
Hybrid Quantum Neural Network for Phase Recognition
Paper Summary
Title: Hybrid quantum-classical neural network for sample-efficient recognition of topological phases
arXiv: 2606.28199 (and companion paper 2606.28201)
Authors: Markus K. Hoffmann, Leon C. Sander, Colin Scarato et al. (ETH Zurich / Wallraff group)
Date: June 26, 2026
Core Innovation
A hybrid quantum-classical neural network that combines a shallow parameterized quantum circuit with a classical neural network for sample-efficient recognition of quantum phases. The parameterized quantum circuit performs a nonlocal transformation of the measurement basis, jointly trained with the classical neural network to maximize statistical distance between measurements of different quantum states.
Key Results
Sample Efficiency
- Training sample complexity: Reduced by ~10x compared to classical neural network on randomized Pauli measurements
- Inference sample complexity: Also reduced by ~10x
- Shallow quantum circuit: Compatible with existing quantum computers (no deep circuits needed)
Phase Recognition Performance
- Topological phase recognition: Distinguishes surface code topological phase from symmetry-enriched topological phase and random product states
- Single-shot accuracy: >85%
- Averaged accuracy (10 measurements): >99%
- Error resilience: Distinguishes topological states even with single-qubit Pauli errors
Hardware Implementation
- Platform: Superconducting quantum hardware
- System size: Surface code lattices up to 4x4 sites in magnetic field
- Real hardware loop: Joint training with actual quantum processor in optimization loop
Technical Architecture
Hybrid Neural Network Components
- Parameterized Quantum Circuit (PQC): Shallow circuit performing nonlocal measurement basis transformation
- Measurement layer: Projects quantum state to classical data
- Classical Neural Network: Feedforward network processing measurement outcomes
Training Process
- Joint optimization: PQC parameters and classical NN weights trained together
- Objective: Maximize statistical distance between data from different quantum states
- Supervised learning: Labeled examples from known quantum phases
Advantages Over Pure Classical Approach
- Nonlocal feature extraction: Quantum circuit captures nonlocal correlations inaccessible to local measurements
- Reduced measurement overhead: Fewer measurements needed for same accuracy
- Shallow circuit: No need for deep quantum circuits that exceed coherence times
When to Use
- Characterizing quantum states on near-term quantum devices
- Recognizing topological phases of matter
- Scenarios where classical methods have unfavorable sample complexity scaling
- Experimental quantum state characterization
- Distinguishing topological from trivial phases
Implementation Considerations
- Quantum circuit must be shallow enough for hardware coherence times
- Statistical distance metric choice affects training stability
- Measurement basis optimization is key to sample efficiency
- Joint classical-quantual training requires hardware-in-the-loop
Pitfalls
- Requires access to quantum hardware for training
- Shallow circuit limits expressivity
- Measurement shot noise affects training convergence
- Scaling to larger systems needs further validation
Related Skills
hybrid-quantum-classical-nn - General hybrid QNN patternsunsupervised-quantum-ml-phase-detection - Unsupervised phase detectionqml-expressivity-trainability - QNN expressivity analysisquantum-neural-measurement-dynamics - Measurement dynamics in QNNs
References
- arXiv:2606.28199 - Sample-efficient phase recognition
- arXiv:2606.28201 - Experimental demonstration on superconducting hardware
