In the context of this project, I am exploring various techniques to enhance the transparency and interpretability of machine learning models. These methods fall into two main categories: auto-explanation and post hoc explanation, with the aim of providing clearer insights into the model's decision-making processes.
Explainable Artificial Intelligence (XAI) aims to make machine learning models more transparent and understandable for humans. It includes two main approaches: auto-explanation, where classifiers generate their own explanations, and post hoc explanation, which separates prediction and explanation steps. XAI considers agnosticism assumptions to determine available knowledge for explanation, with model-agnostic and data-agnostic methods like LORE and Growing Spheres offering flexibility while respecting data privacy.
Explanations can be global, providing insights into the overall behavior of the classifier, or local, focusing on specific predictions. However, there are trade-offs, such as between flexibility and the relevance of explanations. XAI contributes to a better understanding of AI model decisions.