Quantum Machine Learning

Quantum machine learning uses parameterized quantum circuits as trainable models -- variational classifiers, quantum kernels, and quantum neural networks -- trained in a hybrid quantum-classical loop.

The landscape

Quantum machine learning (QML) covers models whose core is a quantum circuit. Data is encoded into qubits (amplitude, angle, or basis encoding), a parameterized circuit transforms it, and measurements produce a prediction. Parameters are trained classically, usually with the parameter-shift rule for gradients.

Main families

• Quantum kernel methods -- a quantum feature map feeding a classical SVM.
• Variational classifiers and quantum neural networks -- trainable circuits as the model itself.
• Generative models such as Born machines and quantum GANs.

Promise versus practice

QML is genuinely expressive, but three obstacles bite: barren plateaus (gradients that vanish exponentially with qubit count), heavy measurement cost, and the absence of a proven general advantage over strong classical ML. Most demonstrations tie, not beat, their classical baselines.

What is real today

What is real now is the tooling: small, honest, runnable QML pipelines on real data, hardware-ready via one code path. Kronos ships several (KQERN quantum kernels, KQUBIT VQE) and reports their results without inflation. The honest timeline stands: a quantum-ML advantage for fusion is a post-~2036 prospect, not a present-day speedup.