Quantum Machine Learning

When quantum meets neural networks

Quantum machine learning explores how quantum computers can accelerate or enhance machine learning tasks. This module covers variational quantum circuits, quantum kernels, and the current state of the field.

Variational Quantum Circuits

Variational Quantum Circuits (VQCs) are parameterized quantum circuits optimized using classical gradient descent. They are the quantum analog of neural networks — layers of gates with tunable parameters.

A VQC consists of three parts: data encoding (loading classical data into quantum states), a parameterized circuit (the quantum model), and measurement (extracting classical information).

•Data encoding: Map classical data to quantum amplitudes or angles.
•Ansatz: A parameterized circuit whose gates depend on trainable parameters.
•Measurement: Expectation values of observables serve as model outputs.
•Hybrid: Classical optimizer updates parameters based on measured gradients.

Quantum Kernels

Quantum kernel methods use a quantum computer to compute similarity measures (kernels) between data points in a high-dimensional quantum feature space. The kernel trick allows learning in this space without explicit feature mapping.

Quantum kernels can access feature spaces exponentially larger than classical kernels, potentially enabling better classification boundaries for certain datasets.

K(x, x') = |\langle \phi(x) | \phi(x') \rangle|^2

Quantum kernel

|\phi(x)\rangle = U(x) |0\rangle^{\otimes n}

Feature map

Current Challenges and Promise

QML is an active research area with both promise and challenges. Barren plateaus — exponentially vanishing gradients in large circuits — make training difficult. Data loading can erase quantum advantages.

However, QML may excel at specific tasks: learning from quantum data, certain structured datasets, and tasks where quantum feature spaces offer natural advantages. The field is evolving rapidly.

•Barren plateaus: Gradients vanish exponentially with circuit depth and qubit count.
•Data loading: Encoding classical data can cost as much as the quantum speedup.
•Quantum data: QML naturally processes data from quantum sensors and experiments.
•NISQ era: Current devices can run small QML models on toy problems.

Try It Yourself

Run this code in your own environment to experiment with the concepts.

PennyLane: Quantum Neural Network

A simple variational classifier for binary classification.

qnn.py

import pennylane as qml
from pennylane import numpy as np

dev = qml.device('default.qubit', wires=2)

def layer(params):
    qml.RX(params[0], wires=0)
    qml.RY(params[1], wires=1)
    qml.CNOT(wires=[0, 1])

@qml.qnode(dev)
def qnn(params, x):
    qml.RY(x[0], wires=0)
    qml.RY(x[1], wires=1)
    for p in params:
        layer(p)
    return qml.expval(qml.PauliZ(0))

# Training would go here
params = [np.array([0.1, 0.2]) for _ in range(2)]
x = np.array([0.5, 0.3])
print('QNN output:', qnn(params, x))

Self-Check Quiz

1. What is the main challenge when training deep variational quantum circuits?

2. What does a quantum kernel compute?

3. What type of data might QML naturally excel at processing?

Quantum Simulation

import pennylane as qml from pennylane import numpy as np dev = qml.device('default.qubit', wires=2) def layer(params): qml.RX(params[0], wires=0) qml.RY(params[1], wires=1) qml.CNOT(wires=[0, 1]) @qml.qnode(dev) def qnn(params, x): qml.RY(x[0], wires=0) qml.RY(x[1], wires=1) for p in params: layer(p) return qml.expval(qml.PauliZ(0)) # Training would go here params = [np.array([0.1, 0.2]) for _ in range(2)] x = np.array([0.5, 0.3]) print('QNN output:', qnn(params, x))