Quantum Neural Networks QNN - RPIQuantumComputing/QuantumCircuits GitHub Wiki

Quantum Neural Networks is an extension of its classical counterparts into quantum mechanics, it focuses on adapting classical neural networks with quantum mechanics in order to incorporate quantum advantages. The incorporation allows for the neural networks to exponentially decrease the time it takes for the training of the current neural network and it could perhaps help scientists and researchers better understand the human consciousness.

To begin, a classical neural network, biologically speaking is just a group of neutrons all connected by synapses, and artificially speaking just a group of nodes, or artificial neutrons, connected to each other in order to create a wide range of abilities as expected of something inspired by the human brain. They are designed to process grid like objects, some examples include images and pixel art, and they have the ability of image classification, object detection, and image segmentation. The neurons are the basic building blocks of the neural network, they receive input signals, perform computations, and produce output signals. The neurons are separated into different layers within the network, some of the most common layers include an input layer, a hidden layer, and an output layer, these layers perform tasks based on their names with the input layer receiving the data and the output layer outputting the result, the hidden layer is any intermediate layer between the input and output layers and they help the network learn complex relationships and extract information from the input, there can also be more than one hidden layer.

Quantum Neural Networks, or QNN, is not just an extension of classical neural networks, it is more an reimagining combined with quantum mechanics principles. Quantum neural networks and classical neural networks are actually very similar in that they both uses the same layered structure, with quantum neural networks incorporates quantum mechanics into its layers: quantum input layer, where it prepares the quantum states for representing the data, encoding the classical data into quantum states using q-bits which are the subsequent inputs to the second layer, the second layer is called the quantum state layer, where like its classical counterpart is where operations on the inputted information takes place, however, unlike classical hidden layers, the quantum gate layer is used to transform the quantum states through specific quantum operations such as enabling the QNN to manipulate the information encoded through transformation and rotation, the layer can also take advantage of the unique properties of quantum mechanics and process multiple states at once, enabling the ability to analyze and go through large amounts of data, where a classical machine may have to sacrifice resolution to go through, at once. The layer can also use quantum entanglement create complex quantum states to identify some relationships and dependencies between the quantum states, the layer can also perform non-linear transformation about the quantum states in order to determine more complex patterns and perform more intricate computations. Finally, the third layer, the measurement layer, is the output layer of the QNN, after all the quantum computation have been applied, the quantum state is then measured to obtain the classical data outputs, this layer basically transforms the quantum states back into classical data to be analyzed, this layer is necessary because classical data are much more easier to process than its quantum counterparts.

On paper, Quantum Neural Networks seems much better than classical neural networks, however, there are several disadvantages of QNN that limits its development: QNN is limited by hardware as the current state of quantum technology as of 2023 presents several challenges to implementing QNNs, some results of this include limited q-bit coherence and high error rates, and error corrected quantum hardware components aren't readily available. Another error is the complexity of the training, since quantum neural networks are still under development and there are still many challenges and problems involved in the machine learning part of the quantum machine learning. Another important issue is the fact that quantum neural networks aren't applicable in every circumstance and may not provide an advantage over classical neural networks in every application as the potential benefits of QNNs are more evident for only certain types of problems, some examples include optimization, quantum chemistry simulations, and quantum data analysis, but for other types of problems that do not exhibit significant advantage for quantum neural networks, classical neural networks may still be more effective and efficient until quantum computing becomes more advanced.