Operational Explications of Classifier Predictions ((11) A05)

我们为机器学习技术开发了一个解释框架，特别关注神经网络。由于目前的分类器非常脆弱，我们将致力于分类器决策的可证明鲁棒性。对于这样的强大的分类器，我们将提供操作解释，向用户呈现输入将如何改变，以便分类器改变其决定，优先考虑输入中的轻微变化。我们将调查有效的可视化和演示的结果，旨在在一个交互式的过程中探索操作的解释在附近的原始输入，支持用户的分类器决策，并支持专家在机器学习时调试。

We develop an explication framework for machine learning techniques with a particular focus on neural networks. As current classifiers are extremely brittle, we will work on provable robustness of the classifier decisions. For such robust classifiers we will provide operational explications presenting to the user how the input would have to change so that the classifier changes its decision, giving priority to slight changes in input. We will investigate efficient visualisation and presentation of the results, aiming at an interactive process exploring operational explications in the neighbourhood of the original input, supporting users in plausibilising the classifier decision, and supporting experts in machine learning when debugging.