Faithfulness Metric

Faithfulness metrics aim to evaluate how accurately explanations of machine learning models reflect the models' actual decision-making processes. Current research focuses on developing and comparing these metrics across various model types, including large language models, graph neural networks, and vision transformers, often within the context of specific tasks like summarization and image generation. The lack of consistent agreement among existing metrics highlights a critical need for improved evaluation methodologies, impacting the reliability of model interpretability and the development of trustworthy AI systems. This work is crucial for building more transparent and accountable AI applications.