论文阅读 [TPAMI-2022] Building and Interpreting Deep Similarity Models

论文阅读 [TPAMI-2022] Building and Interpreting Deep Similarity Models

论文搜索(studyai.com)

搜索论文: Building and Interpreting Deep Similarity Models

搜索论文: http://www.studyai.com/search/whole-site/?q=Building+and+Interpreting+Deep+Similarity+Models

关键字(Keywords)

Machine learning; Data models; Robustness; Neural networks; Taylor series; Feature extraction; Deep learning; Similarity; layer-wise relevance propagation; deep neural networks; explainable machine learning; digital humanities

机器学习

异常检测; 最近邻; 聚类

摘要(Abstract)

Many learning algorithms such as kernel machines, nearest neighbors, clustering, or anomaly detection, are based on distances or similarities.

许多学习算法，如核机器、最近邻、聚类或异常检测，都是基于距离或相似性的。.

Before similarities are used for training an actual machine learning model, we would like to verify that they are bound to meaningful patterns in the data.

在将相似性用于训练实际的机器学习模型之前，我们希望验证它们是否与数据中有意义的模式相关联。.

In this paper, we propose to make similarities interpretable by augmenting them with an explanation.

在本文中，我们建议通过增加解释来解释相似性。.

We develop BiLRP, a scalable and theoretically founded method to systematically decompose the output of an already trained deep similarity model on pairs of input features.

我们开发了BiLRP，这是一种可扩展的、有理论基础的方法，可以系统地分解已经训练好的深度相似模型在成对输入特征上的输出。.

Our method can be expressed as a composition of LRP explanations, which were shown in previous works to scale to highly nonlinear models.

我们的方法可以表示为LRP解释的组合，这些解释在以前的工作中已显示为可扩展到高度非线性模型。.

Through an extensive set of experiments, we demonstrate that BiLRP robustly explains complex similarity models, e.g., built on VGG-16 deep neural network features.

通过大量实验，我们证明BiLRP能够有力地解释复杂的相似性模型，例如，基于VGG-16深度神经网络特征的模型。.

Additionally, we apply our method to an open problem in digital humanities: detailed assessment of similarity between historical documents, such as astronomical tables.

此外，我们将我们的方法应用于数字人文学科中的一个开放问题：对历史文献（如天文表）之间的相似性进行详细评估。.

Here again, BiLRP provides insight and brings verifiability into a highly engineered and problem-specific similarity model…

在这里，BiLRP再次提供了洞察力，并将可验证性引入到一个高度工程化、针对特定问题的相似性模型中。。.

作者(Authors)

[‘Oliver Eberle’, ‘Jochen Büttner’, ‘Florian Kräutli’, ‘Klaus-Robert Müller’, ‘Matteo Valleriani’, ‘Grégoire Montavon’]