An awesome research paper published in Arxiv this week by Seldon Data Scientists Arnaud Van Looveren and Janis Klaise titled "Interpretable Counterfactual Explanations Guided by Prototypes". This paper dives into the concept of counterfactuals, which is an ML local model explanation technique that allows you to ask the question "for this ML prediction, what could be the smallest changes I could do to the input to change the outcome?". Being such a computationally expensive task, this paper proposas a new approach to reduce the computational resources required to use this technique.
|