The ultimate goal of quantum computing is to build a fault-tolerant quantum computer (FTQC), where the number of qubits is sufficiently large and the error rate is satisfactorily low, such that quantum algorithms concatenating quantum error correction can outperform classical computers on certain computing tasks. In the middle way towards FTQC, there is an era of intermediate-scale quantum computing (NISQ) where error correction fails due to the error threshold and insufficient number of qubits. However, it is believed that NISQ devices can perform computations with practical uses and quantum advantage on NISQ devices have already been experimentally demonstrated. To make NISQ devices being useful, attempts in several aspects have been making, such as short-depth quantum algorithms, encoding a quantum simulation problem with smaller number of qubits and quantum error mitigation etc. In this talk, the author focuses on error mitigation. In a nutshell, error mitigation performs computations on variant circuits and then use post process to estimate the computation result with a relative low error. The post process can be either coherent such as that in virtual distillation, or classical such as those in error extrapolation, probabilistic error cancellation (PEC) and subspace expansion.
Learning-based quantum error mitigation can extend the feasibility of PEC to large quantum circuits with correlated errors. Conventional PEC obtains the quasi-probability according to the result of gate-set tomography whose resource cost is unacceptable in large quantum circuit. In Learning-based PEC, we represent the quasi-probability with efficient ansatzes and train them to the optimal. The training data is obtained by substituting general unitary gates with Clifford gates. Our demonstration shows the advantage of learning-based PEC in efficacy and accuracy. Generally, the methodology of optimization can be concatenated with kinds of error mitigation strategies and improve their performance. More importantly, for a quantum circuit with O(N) gates prone to errors, the bias after optimized error mitigation is reduce by a large factor of O(\sqrt N), which indicate the scalability of error mitigation.
Dayue Qin (秦大粤) graduated and received his bachelor degree from school of physics and technology, Wuhan University in 2018. Since then, he is a PhD candidate of Ying Li’s group at graduate School of China Academy of Engineering physics. His academic researches embodied in quantum error mitigation and interests in the aspect of applications of near future quantum hardware.