2018/10/09 by Brandon Carter, Jonas Mueller, Carter, Brandon +5 · 1 citation
Computer Science · Materials Science · #Explainable Artificial Intelligence (XAI) #FOS: Computer and information sciences #Machine Learning (cs.LG) #Machine Learning (stat.ML) #Machine Learning and Data Classification #Machine Learning in Materials Science #Topic Modeling
paper · pdf · doi:10.48550/arxiv.1810.03805
openalex publication_date 2018/10/09 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28
Local explanation frameworks aim to rationalize particular decisions made by\na black-box prediction model. Existing techniques are often restricted to a\nspecific type of predictor or based on input saliency, which may be undesirably\nsensitive to factors unrelated to the model's decision making process. We\ninstead propose sufficient input subsets that identify minimal subsets of\nfeatures whose observed values alone suffice for the same decision to be\nreached, even if all other input feature values are missing. General principles\nthat globally govern a model's decision-making can also be revealed by\nsearching for clusters of such input patterns across many data points. Our\napproach is conceptually straightforward, entirely model-agnostic, simply\nimplemented using instance-wise backward selection, and able to produce more\nconcise rationales than existing techniques. We demonstrate the utility of our\ninterpretation method on various neural network models trained on text, image,\nand genomic data.\n