Low-Depth Fashion-MNIST

Data for benchmarking machine learning models, generated for an upcoming paper: Typical Machine Learning Datasets as Low-Depth Quantum Circuits.

Description of the dataset

The Fashion-MNIST dataset has 28x28 grayscale images of 70,000 fashion items across 10 categories (e.g., T-shirts, trousers, shoes), with 7,000 images per category. Here, we provide circuit parameters that approximate the Flexible Representation of Quantum Images (FRQI) of each image in the Fashion-MNIST dataset.

Additional details

• The class labels are integers from 0 to 9.
• Implementing the circuits in this dataset and obtaining the final state with PennyLane's qml.state() outputs a state vector. This state vector must be processed to recover the original image.
• The dataset contains two circuits per image: those with a depth of four, which are shallower, and those with a depth of eight, which provide more accurate approximations of the exact state.
• The exact_state entry contains a list of numpy arrays representing FRQI states that exactly encode Imagenette images. This significantly increases the file size and can be omitted during download if not needed.

Example usage

import pennylane as qml
import jax

[dataset_params] = qml.data.load("low-depth-fashion-mnist")

def get_circuit(circuit_layout):
    dev = qml.device("default.qubit", wires=11)
    @jax.jit
    @qml.qnode(dev)
    def circuit(params):
        counter = 0
        for gate, wire in circuit_layout:

            if gate == "RY":
                qml.RY(params[counter], wire)
                counter += 1

            elif gate == "CNOT":
                qml.CNOT(wire)

        return qml.state()

    return circuit

# Example for running the circuit with depth 4
circuit_layout_d4 = dataset_params.circuit_layout_d4
circuit_d4 = get_circuit(circuit_layout_d4)
state_d4 = circuit_d4(dataset_params.params_d4[0])

# Example for running the circuit with depth 8
circuit_layout_d8 = dataset_params.circuit_layout_d8
circuit_d8 = get_circuit(circuit_layout_d8)
state_d8 = circuit_d8(dataset_params.params_d8[0])

version 0.1 : initial public release