Counterfactual Explanations via Latent Space Projection and\n Interpolation

Counterfactual explanations represent the minimal change to a data sample\nthat alters its predicted classification, typically from an unfavorable initial\nclass to a desired target class. Counterfactuals help answer questions such as\n"what needs to change for this application to get accepted for a loan?". A\nnumber of recently proposed approaches to counterfactual generation give\nvarying definitions of "plausible" counterfactuals and methods to generate\nthem. However, many of these methods are computationally intensive and provide\nunconvincing explanations. Here we introduce SharpShooter, a method for binary\nclassification that starts by creating a projected version of the input that\nclassifies as the target class. Counterfactual candidates are then generated in\nlatent space on the interpolation line between the input and its projection. We\nthen demonstrate that our framework translates core characteristics of a sample\nto its counterfactual through the use of learned representations. Furthermore,\nwe show that SharpShooter is competitive across common quality metrics on\ntabular and image datasets while being orders of magnitude faster than two\ncomparable methods and excels at measures of realism, making it well-suited for\nhigh velocity machine learning applications which require timely explanations.\n