Learning system, learning method, and storage medium

A learning method, comprising:

• providing a learning system to learn a distribution of a quantum dataset, the learning system including circuitry that includes a quantum computation unit, a classical computation unit, and a management unit, the management unit being configured to control the quantum computation unit and the classical computation unit, the quantum computation unit including an operating unit for a quantum state, a state maintaining unit that maintains the quantum state, and a measuring unit that measures the quantum state;
• determining, by the classical computation unit, a structure of a generator from multiple samples of latent variables and values of quantum circuit parameters;
• calculating, by the quantum computation unit, a ground cost and a gradient of the ground cost from the generator and the quantum dataset;
• calculating, by the classical computation unit, an optimal transport loss and a gradient of the optimal transport loss using the ground cost and the gradient of the ground cost;
• executing, by the classical computation unit, an updating process that updates the quantum circuit parameter using the gradient of the optimum transport loss, thereby reducing the optimum transport loss;
• and
• repeating the updating process by the management unit until the optimum transport loss converges.

A non-transitory computer readable storage medium that stores a program that causes a processor to execute a learning process, wherein the learning process

• determines a structure of a generator from multiple samples of latent variables and values of quantum circuit parameters to learn a distribution of a quantum dataset,
• calculates a ground cost and a gradient of the ground cost from the generator and the quantum dataset,
• calculates an optimal transport loss and a gradient of the optimal transport loss using the ground cost and the gradient of the ground cost,
• executes an updating process that updates the quantum circuit parameter using the gradient of the optimum transport loss, thereby reducing the optimum transport loss, and
• repeats the updating process until the optimum transport loss converges.