Fully Quantum Neural Network
FQNN – Fully Quantum Neural Network
FQNN (Fully Quantum Neural Network) is an innovative neural network architecture. This model relies exclusively on quantum operations, eliminating the need for classical neural layers.
In FQNN, input data is encoded into the quantum states of qubits, and the network parameters (analogous to weights in classical networks) are represented by the rotation angles of quantum gates. These parameters are optimized through the estimation of directional gradients, with the objective function being "fidelity"—a measure of similarity between the network's output state and the state representing the correct class.
The model was tested on the classic Iris dataset using the Qiskit AerSimulator, achieving stable and accurate classification results. These studies confirm the feasibility of building fully quantum classifiers without requiring hybrid architectures that combine classical and quantum elements.
The full description of the architecture and research results was published in the journal Electronics in May 2025:
Ewald, D. The Proposal of a Fully Quantum Neural Network and Fidelity-Driven Training Using Directional Gradients for Multi-Class Classification. Electronics 2025, 14, 2189.
A short explanation in the form of a video: