LAG-XAI: A Lie-Inspired Affine Geometric Framework for Interpretable Paraphrasing in Transformer Latent Spaces

LAG-XAI proposes a geometric framework using Lie algebra-inspired affine transformations to interpret how transformers manipulate text representations in latent spaces. The method provides a structured mathematical lens for understanding transformer behavior during paraphrasing tasks. This contributes to advancing neural network interpretability.