Stop explaining black box machine learning models for high stakes decisions and use interpretable models instead (paper summary)

Posted on December 7, 2021

Deep learning models are usually regarded as black boxes. That is because they are not transparent about the way they reach the prediction. Humans cannot directly interpret the model with millions of parameters. Choosing ignorance can lead to unforeseen dangers. This is inherently a bad practice that should be minimized as much as possible. Current deep learning explainability tools aim to simplify the process to the outcome but does not really explain the “thinking process” that the model followed. ThisRead More