A duplication-free quantum neural network for universal approximation

Abstract

The universality of a quantum neural network refers to its ability to approximate arbitrary functions and is a theoretical guarantee for its effectiveness. A non-universal neural network could fail in completing the machine learning task. One proposal for universality is to encode the quantum data into identical copies of a tensor product, but this will substantially increase the system size and the circuit complexity. To address this problem, we propose a simple design of a duplication-free quantum neural network whose universality can be rigorously proved. Compared with other established proposals, our model requires significantly fewer qubits and a shallower circuit, substantially lowering the resource overhead for implementation. It is also more robust against noise and easier to implement on a near-term device. Simulations show that our model can solve a broad range of classical and quantum learning problems, demonstrating its broad application potential.

15 pages, 10 figures

References in corpus (7)

• Quantum Data Fitting
• Quantum reinforcement learning
• Quantum-state preparation with universal gate decompositions
• Experimental quantum speed-up in reinforcement learning agents
• Simulating a perceptron on a quantum computer
• Near-Term Quantum Computing Techniques: Variational Quantum Algorithms, Error Mitigation, Circuit Compilation, Benchmarking and Classical Simulation
• Nonlinear Quantum Neuron: A Fundamental Building Block for Quantum Neural Networks

Cited by in corpus (3)

• Ensemble-learning error mitigation for variational quantum shallow-circuit classifiers
• Randomness-enhanced expressivity of quantum neural networks
• Measuring Incompatible Observables with Quantum Neural Networks