RISE: Randomized Input Sampling for Explanation of Black-box Models

• DOI: 10.48550/arxiv.1806.07421
• Publication Date: 2018-01-01
• Authors: Vitali Petsiuk; Abir Das; Kate Saenko
• Keywords: Benchmark (surveying); Black box; White box; Deep Neural Networks

Deep neural networks are being used increasingly to automate data analysis and decision making, yet their decision-making process is largely unclear difficult explain the end users. In this paper, we address problem of Explainable AI for deep that take images as input output a class probability. We propose an approach called RISE generates importance map indicating how salient each pixel model's prediction. contrast white-box approaches estimate using gradients or other internal network state, works on black-box models. It estimates empirically by probing model with randomly masked versions image obtaining corresponding outputs. compare our state-of-the-art extraction methods both automatic deletion/insertion metric pointing based human-annotated object segments. Extensive experiments several benchmark datasets show matches exceeds performance methods, including approaches. Project page: http://cs-people.bu.edu/vpetsiuk/rise/