UC Berkeley

Quantum errors remain the primary barrier inhibiting quantum computers from outperforming classical supercomputers. To overcome this challenge, a diverse array of strategies has been developed, encompassing quantum error correction and quantum error mitigation. Machine learning, maturing as a widely adopted approach for pattern recognition, offers new perspectives in enhancing the aforementioned strategies to tackle quantum errors. Furthermore, the implications of quantum errors extend to various applications of quantum computing, notably in quantum machine learning which leverages quantum resources for potential advantage over classical counterparts. This dissertation delves into these intertwined parts, examining the interplay between quantum computation and machine learning. The first part concerns machine learning for enhancing quantum computations. It addresses challenges in correcting errors that occurred to continuously measured logical states, and in improving the efficiency in mitigating errors on both small- and large-scale quantum circuits for increased accuracies in the targeted expectation values, serving as an example of using classical machine learning on quantum data. The second part of this dissertation explores quantum computation for machine learning. It provides theoretical and numerical analysis on the robustness of quantum machine learning models against worst-case errors on input encoded quantum states received through quantum communication, or against quantum decoherence during model training and evaluation, serving as an example of applying quantum machine learning on classical data.