HQNN
Hybrid Quantum Neural Network
Combine quantum layers with classical deep learning to enhance feature expressivity — transparent, reproducible, and benchmarked on Superpositions Studio.
Overview
A Hybrid Quantum Neural Network (HQNN) integrates parameterized quantum circuits as layers within a classical neural network. Classical components (e.g., CNN/MLP) handle robust feature extraction, while quantum layers provide additional feature transformations that may be hard to emulate classically.
Why It Matters
Hybrid models exploit mature classical training infrastructure while exploring potential quantum advantages in representation learning, especially for small and structured datasets.
How HQNN Works
A four-step process to build hybrid quantum-classical neural networks
Build a classical backbone
Build a classical backbone (e.g., CNN or MLP) for initial features.
Insert a quantum layer
Insert a quantum layer (PQC) that transforms features via a feature map and trainable gates.
Optimize end-to-end
Optimize end-to-end with stochastic optimizers; backprop passes through parameter-shift or gradient-free estimators.
Evaluate and ablate
Evaluate calibration, robustness, and ablate the quantum layer's contribution.
Real-World Applications
Where quantum kernels provide tangible advantages
Small/medium datasets
Small/medium datasets with nonlinear decision boundaries
Signals & anomalies
Signals & anomalies (spectra, time series, scientific data)
Research on quantum-enhanced features
Research on quantum-enhanced feature maps/kernels
Hybrid pipelines
Build end-to-end hybrid pipelines that blend PyTorch modules and quantum layers
Strengths & Limitations
Strengths
- Leverages classical strengths while exploring quantum features
- Modular and extensible in standard ML frameworks
- End-to-end training and ablations
Limitations
- No guaranteed advantage; benefits are task-dependent
- Training noise and shot cost can be high
- Careful engineering needed to avoid barren plateaus
Benchmarking and Verification
Benchmark against classical backbones without quantum layers; report deltas in accuracy, calibration, robustness, and compute cost. Seed-controlled pipelines and saved checkpoints ensure reproducibility.
Hardware & Requirements
Proof-of-Concept Example
Real experimental results demonstrating HQNN performance
Task
Time Series Prediction of wind energy
Optimizer
Adam and parameter-shift for quantum part
FAQ
Common questions about HQNN implementation and performance
Explore More Algorithms
Expand your quantum computing capabilities
Ready to Run HQNN?
Run HQNN on Superpositions Studio — plug quantum layers into your DL models and export reproducible experiments.
Powered by Superpositions Studio — Transparent, reproducible quantum computing