Measuring Incompatible Observables with Quantum Neural Networks
The Heisenberg uncertainty principle imposes a fundamental restriction in quantum mechanics, stipulating that measuring one observable completely erases the information on its conjugate one, thereby preventing simultaneous measurements of incompatible observables. Quantum neural networks (QNNs) is one of the most significant applications on near-term devices in noisy intermediate-scale quantum era. Here, we demonstrate that by implementing a multiple-output QNN that emulates a unital quantum channel, one can measure the expectation values of many incompatible observables simultaneously by Pauli-$Z$ measurements on distinct output qubits. We prove the existence of such quantum channel, derive analytical scaling constraints of the measured expectation values, and validate this framework by numerical simulations of observables learning tasks. Notably, our analysis reveals that it requires fewer copies of state when measuring some incompatible observables by the multiple-output QNNs, which demonstrates a resource efficiency advantage compared to separately applying projective measurements.