Quantum Eyes

The ability for an organ to take photons from the outsight world, focus them, and convert them into electrical signals is pure awesomeness! But what’s even more awesome is the organ behind your eyeballs – the brain! The brain is able to take those electrical signals, convert them to images, and then figure out things like, who that person is across the street or what those funny symbols mean

A Machine That Sees

Let’s take the MNIST Dataset, a dataset of digits from 0 to 9: Each digit is a 28 x 28 image, meaning there’s a total of 784 pixels in the whole image.

• Take our image and flatten it, and pop this guy into our neural network and on the other side we end up with our prediction!
• This simple neural network is able to achieve an impressive 88% accuracy!

CNNs

The 3 main parts of a CNN are: Convolutional Layers, Max Pooling Layers and Fully Connected Layers

• CNNs are able to achieve insane accuracy – 99.75% accuracy
• This is due to it’s ability to look at the surrounding pixels and, based off of that, extract features and information from that patch of pixels. Normal NNs can’t do that, so they’re a bit less powerful.

Disadvantages

Many Executions Needed

• Difficult to optimize
• Researchers don’t know which protocols are good, useless, or counter productive
• New Quantum Machine Learning algorithms are already outperforming old ones
• Existing ML algorithms could be translated into QML algorithms, allowing them to take avantage of new properties

Quantum Convoluional Neural Networks

This is a neural network that literally replicates the whole CNN architecture

• Convolutional layers, Pooling Layers, Fully Connected Layer
• The architecture itself looks like this
• A normal MERA takes 1 qubit and then exponentially increases the number of qubits by introducing new qubits into the circuit
• In the reverse MERA, we’re doing the opposite by exponentially decreasing the qubits

CNN’s have 3 main features:

Convolutional layers

• Max Pooling
• Linear Layers
• When you stack layers on top of layers, you get awesome properties.
• A quantum convolutional neural network literally replicates the whole architecture of a CNN, and a Quanvolutional Neural Network is made up of quanvolutionally layers, which replicates convolutionsal layers.

Quanvolutional Neural Networks

A QNN is basically a CNN but with quanvolutionally layers

• Much like a normal convolutional layer, we take a subset of the image, pass it through our kernal, and then it outputs a part of the new image
• We do this over and over again until we reach the linear layer, which classify’s the image

Advantages

Noise Resistant: With Quantum Error Correction, along with it’s quantum nature, QNNs are resistant to constant noise.

Source