← AI Terminology
Quantum Machine Learning
Quantum machine learning (QML) is an interdisciplinary field exploring the intersection of quantum computing and machine learning — investigating whether quantum computers can offer computational advantages for ML tasks through superposition, entanglement, and quantum parallelism.
It remains largely theoretical or pre-commercial as of 2025, awaiting fault-tolerant quantum hardware.
It remains largely theoretical or pre-commercial as of 2025, awaiting fault-tolerant quantum hardware.
Why It Matters in AI
Classical ML faces fundamental computational limits: training large models on classical hardware is bounded by FLOP throughput and memory bandwidth. Quantum computers offer potential exponential speedups for certain linear algebra operations (HHL algorithm) and optimisation problems — theoretically enabling faster kernel methods, PCA, and some types of neural network training. However, quantum advantage for ML has not been demonstrated practically on real problems at scale, and current quantum hardware (NISQ era) is too noisy and small for meaningful QML applications.
Key Points
| Aspect | Description |
|---|---|
| VQC | Variational Quantum Circuit — parameterised quantum circuit trained like a neural network; most practical near-term QML approach |
| NISQ era | Noisy Intermediate-Scale Quantum: current hardware — 100–1,000 noisy qubits, no error correction |
| Timeline | Practical QML advantage for real-world problems: consensus estimate 5–15+ years away |
| Frameworks | PennyLane (Xanadu), Qiskit ML (IBM), TensorFlow Quantum (Google) — hybrid quantum-classical |
| Dequantisation | Many proposed quantum ML speedups can be classically matched with advanced classical algorithms — narrows advantage claims |
| Quantum advantage | Proven theoretically for specific problems (e.g. quantum SVMs, HHL for linear systems) — not yet practical |
Simple Analogy
A revolutionary new type of abacus: theoretically capable of certain calculations exponentially faster, but currently too fragile and error-prone to outperform a modern calculator on any real problem. The promise is real; the practical tools are not yet ready.
Common Usage Examples
- PennyLane:
@qml.qnode(dev); def circuit(params): qml.RX(params[0], wires=0); return qml.expval(qml.PauliZ(0)) - Qiskit ML:
QSVCClassifier(quantum_kernel=kernel)— quantum support vector classifier - TensorFlow Quantum:
tfq.layers.PQC(circuit, observables)— parameterised quantum circuit layer - Variational Quantum Eigensolver (VQE): optimise quantum circuit params — chemical simulation focus
- Hybrid: classical neural network + quantum layer + classical output — most practical current architecture
Summary
In short: Quantum machine learning explores potential quantum speedups for ML tasks — theoretically promising but practically limited by current noisy quantum hardware, with real-world advantage on meaningful ML problems still years away.