QML Architecture

Kinich's quantum machine learning stack provides scikit-learn-style APIs for hybrid quantum-classical models.

Status: The QML module is experimental. Algorithms and APIs are under active development — expect change.

Algorithms

Algorithm	Type	Use case
VQC — Variational Quantum Classifier	Variational classifier	Binary/multiclass classification
QSVM — Quantum SVM	Kernel method	Small-N classification with quantum feature maps
VQNN — Variational QNN	Parameterized circuit	Learned quantum feature transforms

Feature Maps

The choice of feature map determines the quantum feature space:

ZZ Feature Map — entangling; Pauli-Z basis correlations
Pauli Feature Map — X/Y/Z rotation-based encoding
IQP Encoding — instantaneous quantum polynomial circuits
Amplitude / Angle encoding — state-preparation approaches

Optimizers

Optimizer	Character
COBYLA	Gradient-free; noise-resistant
SPSA	Stochastic perturbation — good for noisy hardware
QuantumAdam	Adaptive-moment estimation for quantum parameters

Integration Pattern

Classical data ──▶ Nawal feature extraction
                        │
                        ▼
              Kinich QML model (VQC / QSVM / VQNN)
                        │
                        ▼
              Prediction + classical fallback

Kinich models accept classical feature vectors, encode them into quantum feature spaces, train the parameterized circuits against a loss, and return classical predictions. PyTorch interop (TorchQuantumNeuralNetwork) allows quantum layers inside classical models.

Classical Fallback