The basic algorithmic approach of neural network models for solving the brachisto-chrone problem was analyzed. The Tensor Fowl deep learning framework's neural network module was utilized to numerically solve the brachistochrone problem, with specific solution steps provided and the impact of neural network hyperparameters on the solution results discussed. Unlike traditional neural network applications, this study explored the application of neural network models in variational function fitting, offering references and insights for applying neural networks and other large AI models to university physics teaching, experiments, and research.

A simulation of the Michelson interferometer based on first-principles was conducted to investigate the influencing factors of interference fringes. The simulation results show that angular deflections of the mirror and beam splitter cause the interference fringes to transform between circles, ellipses, and straight lines. When the compensating plate is insufficiently compensated, the interference fringes deviate from circular patterns. In the absence of a compensating plate, the interference fringes transform between ellipses, approximate straight fringes, or hyperbolas as the distance between the two mirrors changes, with two critical points observed. The numerical simulation results are in good agreement with the experimental findings.