Experimenting with Quantum Machine Learning in Qiskit

Table of Contents

1. Introduction
2. Why Use Qiskit for QML?
3. Qiskit Machine Learning Overview
4. Installing and Setting Up Qiskit ML
5. Qiskit Data Encoding Techniques
6. Feature Map Circuits for Classification
7. Variational Quantum Classifiers (VQC)
8. Building a Simple Quantum Classifier
9. Training and Evaluation
10. Using Quantum Kernels with SVM
11. Multiclass Classification Strategies
12. Regression with Quantum Circuits
13. Hardware-Aware Simulation in Qiskit Aer
14. Running QML Models on Real IBM Quantum Hardware
15. Integrating Classical Preprocessing with QML
16. Visualizing Quantum Decision Boundaries
17. Parameter Shift Gradients and Optimization
18. Challenges and Best Practices
19. Applications and Case Studies
20. Conclusion

1. Introduction

Qiskit is IBM’s open-source quantum computing SDK. Its qiskit-machine-learning module provides tools to build, train, and evaluate quantum machine learning models using simulators or real quantum hardware.

2. Why Use Qiskit for QML?

• Direct access to IBM QPUs
• Integration with Qiskit Terra and Aer
• Native support for quantum feature maps, VQCs, and kernels
• Strong documentation and community support

3. Qiskit Machine Learning Overview

• Core components include quantum classifiers, regressors, and kernel-based learners
• Seamless integration with NumPy, SciKit-learn, and Qiskit Aer backends

4. Installing and Setting Up Qiskit ML

pip install qiskit qiskit-machine-learning

5. Qiskit Data Encoding Techniques

• Feature maps transform classical data into quantum states
• Common encodings: ZZFeatureMap, PauliFeatureMap, ZFeatureMap

6. Feature Map Circuits for Classification

from qiskit.circuit.library import ZZFeatureMap
feature_map = ZZFeatureMap(feature_dimension=2, reps=2)

7. Variational Quantum Classifiers (VQC)

• Learn trainable quantum parameters to minimize classification loss
• Combine feature map with variational ansatz

8. Building a Simple Quantum Classifier

from qiskit_machine_learning.algorithms import VQC
from qiskit_machine_learning.circuit.library import RawFeatureVector
from qiskit.circuit.library import TwoLocal

ansatz = TwoLocal(2, ['ry', 'rz'], 'cz', reps=3)
vqc = VQC(feature_map=feature_map, ansatz=ansatz, optimizer='SPSA')

9. Training and Evaluation

• Use datasets like Iris, Breast Cancer, or synthetic XOR
• Split data and train using .fit() and .score() methods

10. Using Quantum Kernels with SVM

from qiskit_machine_learning.kernels import QuantumKernel
from sklearn.svm import SVC

qkernel = QuantumKernel(feature_map=feature_map)
kernel_matrix = qkernel.evaluate(x_train, x_train)
svc = SVC(kernel='precomputed').fit(kernel_matrix, y_train)

11. Multiclass Classification Strategies

• One-vs-Rest or One-vs-One approaches with VQC or Quantum SVM
• Wrap QML model inside sklearn.multiclass.OneVsRestClassifier

12. Regression with Quantum Circuits

• Use qiskit_machine_learning.algorithms.VQR for variational quantum regression

13. Hardware-Aware Simulation in Qiskit Aer

from qiskit_aer import AerSimulator
sim = AerSimulator(noise_model=noise, method='statevector')

14. Running QML Models on Real IBM Quantum Hardware

• Log in to IBMQ:

from qiskit_ibm_provider import IBMProvider
provider = IBMProvider()
backend = provider.get_backend("ibmq_qasm_simulator")

15. Integrating Classical Preprocessing with QML

• Standardize or normalize input features
• Combine with PCA or feature selection before encoding

16. Visualizing Quantum Decision Boundaries

• Plot fidelity heatmaps or measurement probability landscapes
• Use matplotlib and Qiskit circuit sampling

17. Parameter Shift Gradients and Optimization

• Support for analytic gradients using parameter-shift rule
• Optimizers: SPSA, COBYLA, L-BFGS-B

18. Challenges and Best Practices

• Qubit count and circuit depth affect accuracy
• Use shallow ansatz for NISQ devices
• Avoid overfitting via regularization or shot averaging

19. Applications and Case Studies

• Quantum-enhanced finance and healthcare prediction
• Quantum kernel for molecule property classification

20. Conclusion

Qiskit offers a versatile and accessible platform for experimenting with quantum machine learning. From simulators to real hardware, its tools support rapid prototyping and evaluation of quantum-enhanced models across a wide range of domains.