The discovery of a quantum neural network has been established with the successful run of a specialized neural network on a quantum computer. Researchers at Italy’s University of Pavia have modified a rudimentary artificial neural network that is able to run on a quantum computer.

 

Though still simple in architecture to some of neural networks out in the industry, this proof of concept does demonstrate how neural networks can be run on quantum computers without treating each individual quantum bits as a single neuron.

 

The machine learning algorithm they utilized is something called a perceptron. The perceptron is one of the simplest machine learning algorithm with limited scope and is essentially a single-layer neural network.

 

The algorithm takes an array of numbers as a vector input and multiplies them by a weight vector which produces a single bit output. If the output meets a threshold, it will be a 1 and 0 otherwise.

 

The algorithm itself is not new as it was invented way back in 1957 by a man named Frank Rosenblatt of Cornell Aeronautical Laboratory. Despite this, the algorithm still has some useful applications such as with pixel identification in a digital picture.

 

As the input of pixel values are imputed as an array, the algorithm will have individual weights for each pixel that can be used to determine if the array is a picture of a cat. But the more exciting thing about this is the implications; as this helps develop the foundation for combining quantum computers with artificial intelligence.

 

The research team ran the algorithm on IBM’s cloud based quantum computer called Q Experience, which has the power of 5 qubits. The rudimentary quantum AI conducted an image classification task similar to the aforementioned example. The program can only identify three basic patterns of images as the moment.

 

But the advantage of an ai with quantum computing powers is that it has an exponential increase in the dimension that it can calculate. To put it into  perspective, a perceptron on a classical computer can take an input of X number of dimensions, the quantum version can take 2^X number of dimensions.

 

The single layer perceptron can process straight lines but as we have seen with various applications and advancements in ai, it has the ability to vastly outpace humans. As development progresses, we may begin to see our best ai algorithms running on universal quantum computers, catapulting human progress to a post singularity era.