We formulate a causal extension to the recently introduced paradigm ofinstance-wise feature selection to explain black-box visual classifiers . Ourmethod selects a subset of input features that has the greatest causal effect on the models output . The resulting causal selections are sparser and coversalient objects in the scene . We show the efficacy of our approach on multiple datasets by measuring the post-hoc accuracy and Average Causal Effect of selected features on the model’s output . We quantify the causal influence of a . subset of featuresby the Relative Entropy Distance measure .

Author(s) : Pranoy Panda, Sai Srinivas Kancheti, Vineeth N Balasubramanian

Links : PDF - Abstract

Code :
Coursera

Keywords : causal - model - subset - wise - feature -

Leave a Reply

Your email address will not be published. Required fields are marked *