System and method for loss function metalearning for faster, more accurate training, and smaller datasets

This invention describes a way to automatically create and improve the mathematical formulas, called loss functions, that guide how artificial intelligence models learn. It uses a two-step process: first, a genetic algorithm generates many simple candidate loss functions as tree structures with a maximum depth of two. Then, it fine-tunes the numerical values within the best of these simple formulas to make them even more effective for training AI models.