Application of explainable machine learning to eXtreme Gradient Boosting models of all-optical logic gates

Machine learning (ML) has been widely used in the field of optical information processing. Some ML models have been developed to implement optical fiber communication systems, compensate modulation nonlinearity, model all-optical matching systems, etc. However, the current ML models lack explainability, and the explainability of the ML models becomes a very important research direction. We used the eXtreme Gradient Boosting (XGBoost) to construct ML models for all-optical AND, XOR, NOT logic gates. Then, the Shapley Additive exPlanations (SHAP) is used to explain these ML models. Through explanation, not only the importance of features can be analyzed but also the correlation between the input features of the ML models and the predicted outputs can be provided. Thus, the decision-making process of the ML models is obtained, which helps to optimize the performance of the all-optical logic gates, such as achieving more efficient output power.