Influence Analysis

Influence analysis seeks to quantify how individual data points or components impact the behavior of a machine learning model or a complex system. Current research focuses on developing and refining methods to estimate this influence, employing techniques like influence functions and graph convolutional networks, particularly within the contexts of federated learning, time series analysis, and large language models. These advancements enhance model interpretability, improve model robustness and generalization, and offer insights into the dynamics of complex systems, with applications ranging from fairness assessment to financial prediction.