期刊
IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE
卷 43, 期 10, 页码 3416-3431出版社
IEEE COMPUTER SOC
DOI: 10.1109/TPAMI.2020.2982882
关键词
Visualization; Semantics; Neural networks; Task analysis; Feature extraction; Annotations; Benchmark testing; Convolutional neural networks; interpretable deep learning
资金
- National Natural Science Foundation of China [U19B2043, 61906120]
- DARPA XAI Award [N66001-17-2-4029]
- NSF [IIS 1423305]
- ARO [W911NF1810296]
- U.S. Department of Defense (DOD) [W911NF1810296] Funding Source: U.S. Department of Defense (DOD)
This paper introduces a generic method to learn interpretable convolutional filters in a deep CNN without additional annotations, using the same training data as traditional CNNs. The experiments demonstrate that the interpretable filters are more semantically meaningful than traditional filters.
This paper proposes a generic method to learn interpretable convolutional filters in a deep convolutional neural network (CNN) for object classification, where each interpretable filter encodes features of a specific object part. Our method does not require additional annotations of object parts or textures for supervision. Instead, we use the same training data as traditional CNNs. Our method automatically assigns each interpretable filter in a high conv-layer with an object part of a certain category during the learning process. Such explicit knowledge representations in conv-layers of the CNN help people clarify the logic encoded in the CNN, i.e., answering what patterns the CNN extracts from an input image and uses for prediction. We have tested our method using different benchmark CNNs with various architectures to demonstrate the broad applicability of our method. Experiments have shown that our interpretable filters are much more semantically meaningful than traditional filters.
作者
我是这篇论文的作者
点击您的名字以认领此论文并将其添加到您的个人资料中。
推荐
暂无数据