Quantum kernel classifier

Quantum Machine Learning · course presentation

Classifying data with a quantum kernel

A support vector machine cannot split two rings with a straight line. Let a two-qubit circuit measure how similar each pair of points is, and the same algorithm draws the right boundary.

Download the slides (.pptx) PDF Try the demo

Live demo

Everything below runs in this page. The two-qubit circuit is simulated exactly, so there is no noise.

Class A Class B Test point Wrong
Data
Kernel
Noise 0.06
Training accuracy
–
Test accuracy
–
Support vectors
–
Circuit pairs
–

The slides

Ten slides. Use the arrow keys or the buttons.

Slide 1 of 10
1 / 10

How it works

  1. 1

    Encode

    A circuit of Hadamard, phase and ZZ gates turns a point's two features into a two-qubit state.

  2. 2

    Compare

    Encode one point, undo the encoding of another, and measure. The chance of reading 00 is the similarity of the pair.

  3. 3

    Train

    Fill a matrix with those similarities and give it to an ordinary support vector machine.

This example does not show a quantum advantage. A classical RBF kernel does as well or better on these toy datasets, and two qubits are trivial to simulate. It shows the method.

The numbers on the slides come from experiment.js in the repository and can be reproduced with node experiment.js.