在深度神经网络中预先修剪 Clever-Hans 策略（英文版）

英文标题：Preemptively Pruning Clever-Hans Strategies in Deep Neural Networks中文摘要：本文探讨了解释性人工智能在验证机器学习模型中的应用，发现只有用户接受了解释并不能保证模型的准确性。作者还提供了一种新的方法，即 Explanation-Guided Exposure Minimization (EGEM)，通过预先剪枝未被积极解释反馈的 ML 模型变化来减少对隐藏的 Clever Hans 策略的依赖，最终在自然图像数据上获得更高的预测准确度。英文摘要：Explainable AI has become a popular tool for validating machine learningmodels. Mismatches between the explained model's decision strategy and theuser's domain knowledge (e.g. Clever Hans effects) have also been recognized asa starting