Towards interpreting ML-based automated malware detection models: a survey.

摘要: Malware is being increasingly threatening and malware detectors based on traditional signature-based analysis are no longer suitable for current detection. Recently, the models machine learning (ML) developed predicting unknown variants saving human strength. However, most of existing ML black-box, which made their pre-diction results undependable, therefore need further interpretation in order to be effectively deployed wild. This paper aims examine categorize researches ML-based detector interpretability. We first give a detailed comparison over previous work common model inter-pretability groups after introducing principles, attributes, evaluation indi-cators taxonomy Then we investigate methods towards detection, by addressing importance interpreting detectors, challenges faced this field, solutions migitating these challenges, new classifying all state-of-the-art detection interpretability recent years. The highlight our survey providing interpreta-tion summarized re-searches field. In addition, evaluate approaches method attributes generate final score so as insight quantifying By concluding researches, hope can provide suggestions researchers who interested de-tection models.