Quantum Algorithms for Deep Convolutional Neural Networks
Description
This repository contains an implementation of the quantum convolutional layer and its application to the MNIST classification task in :
- Paper : Quantum Algorithms for Deep Convolutional Neural Networks
- Authors : Kerenidis, Landman and Prakash
- Date : 2019
Setup
To install the quantum convolutional layer, clone this repository and execute the following commands :
$ cd qml-qcnn
$ pip install -e .
Details
The quantum convolutional layer is a PyTorch module with the following parameters :
- in_channels (int): number of input channels.
- out_channels (int): number of output channels.
- kernel_size (int): size of the convolution kernel.
- eps (float): precision of quantum multiplication.
- cap (float): value for cap 'relu' activation function.
- ratio (float): precision of quantum tomography.
- delta (float): precision of quantum gradient estimation.
- stride (int, optional): convolution stride. Defaults to 1.
- padding (int, optional): convolution padding. Defaults to 0.
- dilation (int, optional): convolution dilation. Defaults to 1.
The bias is unsupported for now.
Example
In Pytorch, you can use the 2D convolution module as follows :
from torch import nn
torch_conv2d = nn.Conv2d(16, 32, 3, stride=1)Similarly, you can use the 2D quantum convolution module as follows :
import qcnn
eps = 0.01
cap = 10.
ratio = 0.5
delta = 0.01
quantum_conv2d = qcnn.QuantumConv2d(16, 32, 3, eps, cap, ratio, delta, stride=1)Experiments
The experiments in the paper are reproduced for MNIST classification task using PyTorch Lightning.
- Option 1 : Open in Colab. You can activate the GPU in Notebook Settings.
- Option 2 : Run on local machine. First, you need to install PyTorch Lightning :
$ pip install pytorch_lightning
You can run an experiment using the following command :
$ cd qml-qcnn/scripts/
$ python mnist_classification --eps=0.01 --cap=10 --ratio=0.5 --delta=0.01