QNN
Quantum Neural Networks
Parameterized quantum circuits for classification/regression on NISQ devices — transparent, reproducible, and benchmarked on Superpositions Studio.
Overview
A Quantum Neural Network (QNN) uses parameterized quantum circuits (PQCs) to learn mappings from inputs to outputs. Data are embedded into quantum states via encoders; trainable gates with parameters θ are optimized to minimize a loss function computed from measured observables. QNNs are researched as quantum analogs of neural networks with potentially rich expressivity.
Why It Matters
QNNs explore whether quantum circuits can learn useful representations, especially in small-data regimes and for problems with structure that quantum circuits can exploit. They are a core path for quantum ML research on current devices.
How QNN Works
A five-step process to build and train quantum neural networks for classification and regression
Choose a data encoder
Choose a data encoder to map classical inputs to circuits (angle or amplitude encoding, optional data reuploading).
Design a parameterized ansatz
Design a parameterized ansatz with trainable gates.
Define a loss
Define a loss (e.g., cross-entropy for classification) based on measured observables.
Optimize parameters
Optimize θ with gradient-free or gradient-based methods (parameter-shift, SPSA).
Validate and test
Validate and test generalization.
Real-World Applications
Where QNNs provide practical solutions for quantum machine learning and classification tasks
Classification/regression
Classification/regression machine learning problems
Anomaly detection
Anomaly detection and signal processing prototypes
Quantum feature learning
Quantum feature learning research
Hybrid pipelines
Hybrid pipelines with classical preprocessing or post-processing
Strengths & Limitations
Strengths
- Flexible modeling with PQCs and hybrid training
- Potentially expressive hypothesis spaces
- Integrates with classical ML stacks
Limitations
- Barren plateaus and vanishing gradients
- High shot cost and noise sensitivity
- Advantage over classical NNs unproven generally; task-dependent
Benchmarking and Verification
Compare against classical baselines (logistic regression, SVM, MLP). Report accuracy, learning curves, and model/run parameters (depth, shots). Seed-controlled runs ensure reproducibility.
Hardware & Requirements
Proof-of-Concept Example
Real experimental results demonstrating QNN performance
Task
Binary classification on a 2D moon dataset
Ansatz
Layered entangling circuit with data reuploading (1–3 layers)
FAQ
Common questions about QNN implementation and performance
Explore More Algorithms
Expand your quantum computing capabilities
Ready to Run QNN?
Run QNN on Superpositions Studio — design circuits, train models, and export reproducible results.
Powered by Superpositions Studio — Transparent, reproducible quantum computing