Optimizing the Loss Function of Zero-DCE for Low-Light Image Enhancement Using Large Language Models: A Study

This paper explores the use of Large Language Models (LLMs) to optimize the training of machine learning models, with a focus on enhancing the loss function of the Zero-DCE algorithm for low-light image enhancement. Conventional image enhancement methods rely on manually designed loss functions, which struggle to simultaneously improve brightness, preserve details, and suppress noise. We leverage the semantic understanding and generative capabilities of LLMs to automatically construct multi-objective loss functions and iteratively refine them through a trainingfeedbackregeneration loop. By providing structured performance feedback to the LLM, it generates new loss functions that progressively improve model performance. Experimental results demonstrate that incorporating LLMs into the optimization of machine learning training offers promising research value and practical applicability.