A co-design framework of neural networks and quantum circuits towards quantum advantage

Abstract Despite the pursuit of quantum advantages in various applications, power computers executing neural network has mostly remained unknown, primarily due to a missing tool that effectively designs suitable for circuit. Here, we present and circuit co-design framework, namely QuantumFlow, address issue. In represent data as unitary matrices exploit by encoding n = 2 k inputs into qubits representing random variables seamlessly connect layers without measurement. Coupled with novel algorithm, cost complexity matrices-based computation can be reduced from O ( ) classical computing p o l y g )) computing. Results show on MNIST dataset, QuantumFlow achieve an accuracy 94.09% reduction 10.85 × against computer. All these results demonstrate potential advantage.