nd no post hoc test was performed. Data availability The ML benchmark Fashion-MNIST is available at https://github.com/zalandoresearch/fashion-mnist. The PASCAL VOC2007 dataset is available at http://host.robots.ox.ac.uk/pascal/VOC/voc2007/. The RGB and hyperspectral data that support the findings of this study are available at https://tudatalib.ulb.tu-darmstadt.de/handle/tudatalib/2278.4 and in the code repository https://codeocean.com/capsule/4559958/tree. The user study is available at https://github.com/ml-research/xil/tree/master/Trust_Study.Code availability The code and a fully runnable capsule to reproduce the figures and results of this article, including pre-trained models, can be
Open resource ↗tudatalib · tudatalib/2278.4 · pdf-raw-page:14 lines:1-41Unverified paper record
Right for the Wrong Scientific Reasons: Revising Deep Networks by Interacting with their Explanations.
arXiv (Cornell University) · 15 Jan 2020
Abstract
Deep neural networks have shown excellent performances in many real-world applications. Unfortunately, they may show Hans-like behavior---making use of confounding factors within datasets---to achieve high performance. In this work we introduce the novel setting of explanatory interactive learning (XIL) and illustrate its benefits on a plant phenotyping research task. XIL adds the scientist into the training loop such that she interactively revises the original model via providing feedback on its explanations. Our experimental results demonstrate that XIL can help avoiding Clever Hans moments in machine and encourages (or discourages, if appropriate) trust into the underlying model.
Plant phenotyping relevance
植物フェノタイピング課題を対象に、説明への科学者の介入で深層モデルを修正する新しい機械学習手法を提案しており、方法開発が中心である。
abstractIn this work we introduce the novel setting of explanatory interactive learning (XIL) and illustrate its benefits on a plant phenotyping research task.
abstractXIL adds the scientist into the training loop such that she interactively revises the original model via providing feedback on its explanations.
Code and data availability
The paper's plant-phenotyping measurements (RGB and hyperspectral images of healthy/Cercospora-inoculated sugar beet leaf discs) are publicly deposited at TU Datalib, and the authors' analysis code, runnable capsule, and pre-trained models are publicly available on Code Ocean. Both are explicitly stated in the Data/Coa
ailable at https://github.com/zalandoresearch/fashion-mnist. The PASCAL VOC2007 dataset is available at http://host.robots.ox.ac.uk/pascal/VOC/voc2007/. The RGB and hyperspectral data that support the findings of this study are available at https://tudatalib.ulb.tu-darmstadt.de/handle/tudatalib/2278.4 and in the code repository https://codeocean.com/capsule/4559958/tree. The user study is available at https://github.com/ml-research/xil/tree/master/Trust_Study.Code availability The code and a fully runnable capsule to reproduce the figures and results of this article, including pre-trained models, can be found at https://codeocean.com/capsule/4559958/tree.Statement of ethical compliance The a
Open resource ↗codeocean · capsule/4559958 · pdf-raw-page:14 lines:1-41This is an automatically classified, unverified record. Curator approval is required before any resource enters the Catalog.