Hybrid Quantum-Classical Neural Networks: How Machines Recognize Quantum Phases

A hybrid quantum-classical neural network learns to identify phases of matter by combining quantum measurements with classical deep learning, enabling machines to detect topological order from experimental data.

Animated Pipeline

Preparation |0> qubits lattice state p Parameterized Quantum Circuit Y_theta rotations CZ_theta gates Z readouts Measurement vector x Classical Neural Network fully-connected output label y
Press Play to see data flow through the pipeline, or Step to advance stage by stage.

Toy Simulation: Phase Transition

Drag the slider to adjust the field strength h and observe how topological entanglement entropy S_topo and stabilizer expectation values change. The dashed line marks the phase transition near h = 0.3.

Topological phase
Trivial phase
h_c ~ 0.3
S_topo (topological entanglement entropy) Stabilizer expectation values

Synthetic data for illustration only. The sharp drop in S_topo signals the transition from topological to trivial phase.

Worked Example: 4-Qubit Readout

Consider a 4-qubit system in a topological state. After the parameterized quantum circuit, we measure the Z-basis expectation values for each qubit. The classical neural network then processes these measurements to classify the phase.

QubitZ expectationWeightContribution
q0+0.820.45+0.37
q1-0.630.32-0.20
q2+0.910.28+0.25
q3-0.740.35-0.26

The weighted sum is +0.16, which after a sigmoid activation gives a probability of 0.54 for the topological phase. The network correctly identifies the state as topological (above the 0.5 threshold).

Enjoy this tool? Build your own with Super