Abstract: The exploration of quantum algorithms that possess quantum advantages is a central topic in quantum computation and quantum information processing. Quantum machine learning is one of the
most promising candidates in this area. Numerous quantum machine learning algorithms have been proposed in the past decade, which achieve quadratic or even exponential speedup over their classical
counterparts. However, enormous quantum computational resources are required to implement such algorithms, which are unaffordable for near-term quantum devices. It remains obscure if there
exists any quantum algorithm that can be efficiently implement on near-term quantum devices with potential quantum advantages. To address this issue, we propose two learning schemes, the
quantum nonlinear classification learning scheme and quantum generative adversarial learning scheme, to benefit classical machine learning and quantum information processing, respectively.
For the former scheme, the main ingredients are variational quantum perceptron (VQP) and a quantum generalization of classical ensemble learning. VQP employs parameterized quantum
circuits to learn an amplitude amplification operation with classical optimization. A stronger nonlinear classifier can be established by combining a set of VQPs using a subsampling method,
which has potential to achieve quadratic speedup in computational complexity compared to its classical counterparts. A favorable result of this scheme is that it can process large-scale data
on current quantum devices. For the latter scheme, the core concept is the reformulation of a given quantum information processing task, e.g., entanglement test and state discrimination, as a
quantum adversarial game. The reformulated quantum adversarial game enables efficient solutions, attributed to that the optimization function is convex and the required number of measurements is
exponentially reduced over conventional methods. This scheme can be efficiently implemented on parameterized quantum circuits with a certain architecture with classical optimization. We conduct
several numerical simulations to validate the performance of the above two schemes.