Generative Modelling of Quantum Processes via Quantum-Probabilistic Information Geometry

摘要: Understanding of quantum mechanical systems depends on the interaction between theory and experiment. With the rise of computers in the late 20th century, theory has come to include numerical simulation of physical systems beyond the reach of analytic techniques. While current state-of-the-art simulations are run on classical computers, quantum computers have been proposed to more efficiently simulate quantum systems [8]. Recently, quantum computers have experimentally surpassed the performance of classical computers on the specialized task of simulating the output of random quantum dynamics [3]. In the coming years it is expected that quantum computers will surpass classical computers on increasingly practical tasks.Many algorithms have been proposed for the task of quantum simulation. A universal algorithm for quantum simulation on quantum computers was developed as early as 1996 [17], where techniques for both closed and open quantum systems were proposed. However, these proposals often require quantum circuits with depths far beyond the reach of today’s Noisy Intermediate-Scale Quantum (NISQ) processors [23]. Variational algorithms offer an alternative approach which can reduce the circuit depth requirements for simulation tasks, making them more amenable to near-term hardware at the cost of requiring classical parameter optimization. In particular, these algorithms take the heuristic perspective that has served machine learning well in recent years by defining a loss function on the samples from a quantum computer, and optimizing the parameters of a quantum circuit to minimize that loss [20, 19].