US Patent 12026487 Method for optimizing program using reinforcement learning

A method for automatically optimizing a program based on reinforcement learning includes (a) receiving an input for a source program, which includes a fixed parameter and variable parameter, (b) generating the source program based on the received input, (c) converting the source program into an object program, (d) executing the converted object program to measure a performance of the executed object program, (e) inputting the variable parameter and the measured performance into a machine learning model, and outputting a variation of the variable parameter, and (f) regenerating a source program reflecting the variation of the variable parameter.