Quantum Machine Learning (QML) is an interdisciplinary field at the intersection of quantum computing and machine learning that develops algorithms leveraging quantum mechanical phenomena — superposition, entanglement, and interference — to process data and train models in ways that may offer advantages over classical approaches for certain problem classes.¹⁾ QML encompasses both running classical ML algorithms on quantum hardware and designing fundamentally new learning algorithms that exploit quantum properties.

• Superposition — qubits exist in multiple states simultaneously, enabling exploration of exponentially large solution spaces in parallel²⁾
• Entanglement — correlations between qubits capture complex data patterns that would require exponentially many classical bits to represent
• Interference — quantum amplitudes can be manipulated to amplify correct solutions and suppress incorrect ones, aiding optimization tasks like hyperparameter tuning

VQCs are hybrid quantum-classical models central to near-term QML. They consist of parameterized quantum gates (e.g., rotations R_X(theta), R_Z(theta)) interleaved with entangling operations.³⁾ The circuit processes data quantumly, measurements produce classical outputs (e.g., expectation values), and a classical optimizer adjusts the gate parameters iteratively — analogous to training neural network weights via backpropagation.

VQCs are used in algorithms like the Variational Quantum Eigensolver (VQE) and Quantum Approximate Optimization Algorithm (QAOA).

Quantum kernel methods map classical data into high-dimensional Hilbert spaces using parameterized quantum circuits (quantum feature maps).⁴⁾ The kernel matrix is computed as the overlap between quantum states |phi(x_i)> and |phi(x_j)>, then fed into classical algorithms like Support Vector Machines. This Quantum Support Vector Machine (QSVM) approach exploits quantum advantage for computing inner products in spaces intractable for classical machines.

QNNs blend classical and quantum layers, using quantum circuits as differentiable components within larger hybrid architectures. They are applicable to classification, regression, and generative tasks.⁵⁾

Quantum Generative Adversarial Networks use quantum circuits as generators to learn and reproduce data distributions, with classical or quantum discriminators providing the adversarial training signal.

Current Noisy Intermediate-Scale Quantum (NISQ) devices support practical hybrid workflows:⁶⁾

• Optimization — portfolio optimization, logistics, and combinatorial problems
• Drug discovery — molecular simulation and protein folding
• Materials science — electronic structure calculations
• Financial modeling — risk assessment and derivative pricing
• Classification — QSVM and VQC-based classifiers for structured data

The key constraint is that NISQ hardware is noisy and limited in qubit count, so hybrid approaches that offload only the hardest computational kernels to quantum processors are most practical.

• PennyLane (Xanadu) — open-source framework for differentiable quantum programming, supporting VQCs, quantum kernels, and NISQ applications across multiple backends⁷⁾
• Qiskit Machine Learning (IBM) — library for QSVM, VQC-based models, and quantum kernels integrated with the Qiskit ecosystem
• Cirq (Google) — quantum computing framework supporting QML research
• Classiq — platform offering QNN and QML algorithm development tools⁸⁾
• Amazon Braket — managed quantum computing service supporting multiple hardware backends

Quantum advantage for ML remains an active research question. Theoretical results suggest exponential speedups for certain structured problems (e.g., quantum principal component analysis, quantum sampling), but empirical demonstrations of advantage over classical ML on practical tasks remain limited.⁹⁾ The consensus is that quantum advantage will emerge first for problems with inherently quantum structure rather than for general-purpose ML tasks like image classification.

• Topological Qubits
• Quantum Computing
• Variational Quantum Algorithms