375 followers
An approach through data perturbation (influence functions) does seem viable for interpretable machine learning. In statistics we did not really understand regression methods very well before the advent of regression diagnostics. https://t.co/736sIYiq1i