Quantum Machine Learning Algorithms
From the course Quantum Computing & AI
Quantum machine learning (QML) combines quantum computing with ML techniques. Here are the main algorithmic approaches.
Variational Quantum Algorithms
The most practical approach for near-term quantum computers:
- Variational Quantum Eigensolver (VQE) — optimize quantum circuits to find ground states
- Quantum Approximate Optimization Algorithm (QAOA) — solve combinatorial optimization problems
- Variational Quantum Classifiers — parameterized quantum circuits for classification
How variational algorithms work:
- Prepare a parameterized quantum circuit (ansatz)
- Measure the quantum state to evaluate a cost function
- Use a classical optimizer to update circuit parameters
- Repeat until convergence
Quantum Kernel Methods
Quantum feature maps for classical ML:
- Map classical data into quantum Hilbert space
- Compute kernel values using quantum circuits
- Use classical SVM or kernel regression with quantum kernels
- Potential advantage when quantum feature maps are hard to simulate classically
Quantum Neural Networks
Neural network analogues on quantum hardware:
- Parameterized Quantum Circuits (PQCs) — quantum gates with trainable parameters
- Quantum Convolutional Networks — quantum analogues of CNNs for image classification
- Quantum Recurrent Networks — sequential processing with quantum states
- Quantum Transformer — attention mechanism adapted for quantum computation
Quantum Generative Models
Generating data with quantum circuits:
- Quantum GANs — quantum generator with classical or quantum discriminator
- Quantum Boltzmann Machines — leverage quantum tunneling for better sampling
- Born Machines — directly sample from quantum probability distributions
Theoretical Speedups
Proven quantum advantages for specific tasks:
- HHL Algorithm — exponential speedup for linear systems (with caveats on data loading)
- Grover's Algorithm — quadratic speedup for unstructured search
- Quantum PCA — exponential speedup for principal component analysis
- Quantum Sampling — provable advantages for sampling from certain distributions
Named Example: Xanadu & IonQ
Xanadu's PennyLane framework has become the de facto standard for quantum ML research, with integrations into PyTorch and TensorFlow that make hybrid quantum-classical training nearly identical to classical ML workflows. IonQ's trapped-ion systems offer notably higher two-qubit gate fidelities (>99.5%) compared to superconducting qubits, making them attractive for quantum kernel methods where gate error accumulation is the primary quality bottleneck.
# Variational quantum classifier with PennyLane
import pennylane as qml
import pennylane.numpy as np
dev = qml.device("default.qubit", wires=4)
@qml.qnode(dev)
def circuit(x, weights):
qml.AngleEmbedding(x, wires=range(4)) # Encode data
qml.BasicEntanglerLayers(weights, wires=range(4)) # Parameterized
return qml.expval(qml.PauliZ(0)) # Measure
# Train with standard gradient descent
opt = qml.GradientDescentOptimizer(stepsize=0.1)
weights = np.random.randn(3, 4)
for step in range(100):
weights = opt.step(lambda w: cost(circuit, X_train, y_train, w), weights)
Common Pitfall: Barren plateaus — vanishing gradients in deep variational quantum circuits — make training QML models with many layers extremely difficult. This is an active research problem. Limit circuit depth to under 15 layers and use local cost functions rather than global ones to avoid this pitfall in practice.
Key Takeaways
- •Variational algorithms are the most practical near-term QML approach
- •Quantum kernels map data into high-dimensional Hilbert space
- •Parameterized quantum circuits are the quantum analogue of neural networks
- •Quantum GANs and Born Machines offer new approaches to generative modeling
- •Theoretical speedups exist for linear systems, search, and PCA
Progress is kept in this browser only.