Identifying the Optimal Step Size for Gradient Descent Using Monte Carlo Simulation in Optimisation of Micro-laser Welding of Optoelectronics

The gradient descent/ascent algorithm is a commonly used algorithm in machine learning. It is also a well-established technique within the response surface methodology. Where it requires a number of experiments to be performed. This makes it a resource intensive technique, both in terms of investment of experimentation materials and time. Additionally, this can be exacerbated if a poor choice is made for the step size. The step size is chosen according to expert opinion, which can leave room for doubt as to whether an appropriate step was chosen. To overcome this issue a more rigorous technique is proposed for selecting the step size using a Monte Carlo simulation. This novel application of using a Monte Carlo simulation to determine the step size was implemented in Java. The techniques' application was used for earlier work which developed a surrogate model to optimize the welding location. The optimization was to reduce the post weld stress in optical fibers used in optoelectronics. The surrogate model came from a design of experiment for micro-laser welding of optoelectronics.