This looks like that: deep learning for interpretable image recognition

By Oscar Li, Alina Jade Barnett, Jonathan Su, Cynthia Rudin. 2018

1,568 citations recorded by the source corpus.

When we are faced with challenging image classification tasks, we often explain our reasoning by dissecting the image, and pointing out prototypical aspects of one class or another. The mounting evidence for each of the classes helps us make our final decision. In this work, we introduce a deep network architecture -- prototypical part network (ProtoPNet), that reasons in a similar way: the network dissects the image by finding prototypical parts, and combines evidence from the prototypes to make a final classification. The model thus reasons in a way that is qualitatively similar to the way ornithologists, physicians, and others would explain to people on how to solve challenging image classification tasks. The network uses only image-level labels for training without any annotations for parts of images. We demonstrate our method on the CUB-200-2011 dataset and the Stanford Cars dataset. Our experiments show that ProtoPNet can achieve comparable accuracy with its analogous non-interpretable counterpart, and when several ProtoPNets are combined into a larger network, it can achieve an accuracy that is on par with some of the best-performing deep models. Moreover, ProtoPNet provides a level of interpretability that is absent in other interpretable deep models.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition, Safety, Robustness & Interpretability

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Computer Vision and Pattern Recognition

In Safety, Robustness & Interpretability

Citations in the corpus

SELFEXPLAIN: A Self-Explaining Architecture for Neural Text Classifiers (2021, background): “A class of inherently interpretable classifiers explain model predictions locally using human-understandable high-level concepts such as prototypes (Melis and Jaakkola, 2018; Chen et al., 2019) and interpretable classes (Koh et al.”
Interpretable Machine Learning: Fundamental Principles and 10 Grand Challenges (2021, background): “The works of Li et al. (2018) and Chen et al. (2019) have been extended in the domain of deep case-based reasoning and deep prototype learning.”
The Twin-System Approach as One Generic Solution for XAI: An Overview of ANN-CBR Twins for Explaining Deep Learning (2019, methodology): “Finally, most previous twin-system work has tended to examine traditional neural networks (i.e., MLPs), rather than more recent DNNs (for rare potential exceptions see [Chen et al., 2018] and [Li et al., 2017]).”
Stop Explaining Black Box Machine Learning Models for High Stakes Decisions and Use Interpretable Models Instead (2018, background): “t the explanations are real, and not posthoc. In a recent attempt to do this, Chen, Li, and colleagues have been building architectures that append a special prototype layer to the end of the network [49, 55]. During training, the prototype layer ﬁnds parts of training images that act as prototypes ”
Please Stop Explaining Black Box Models for High Stakes Decisions (2018, background): “In a recent attempt to do this, Li et al. [2018], Chen et al. [2018a] have been building architectures that append a special prototype layer to the end of the network.”
Ethical Implementation of Artificial Intelligence to Select Embryos in In Vitro Fertilization (2021, background): “One key point that has emerged from past research is that as long as one can design the interpretability metric carefully to match the domain, interpretable models tend not to lose accuracy relative to their black-box counterparts [26, 27].”
A case-based interpretable deep learning model for classification of mass lesions in digital mammography (2021, methodology): “This change is equivalent to changing all uses of AVGPOOL to max in equations (6) and (9); and changing mink to min in equation (7).”
Making deep neural networks right for the right scientific reasons by interacting with their explanations (2020, background): “One should improve the quality and robustness of XAI methods and also explore XIL for interpretable models [32].”
Why Are We Using Black Box Models in AI When We Don’t Need To? A Lesson From An Explainable AI Competition (2019, background): “…vision, where deep neural networks (the most difficult kind of black box model to explain) are the state-of-the-art, we and other scientists (e.g., Chen et al., 2019; Y. Li et al., 2017; L. Li, Liu, Chen, & Rudin, 2018; Ming, Xu, Qu, & Ren, 2019) have found ways to add interpretability…”
Learning Optimized Risk Scores (2016, background): “…promising option to build models that obey constraints related to, for example, interpretability (see e.g. Caruana et al., 2015; Gupta et al., 2016; Chen et al., 2018; Li et al., 2018, where interpretability is addressed through constraints on model form), safety (Amodei et al., 2016), credibility…”
Approximating CNNs with Bag-of-local-Features models works surprisingly well on ImageNet (2019, methodology): “Chen et al. (2018) uses a similar approach but is based on a convolutional architecture to extract class-speciﬁc prototype patches.”
Visual Classification via Description from Large Language Models (2022, background): “Aggregating similarities enables interpretability, as the class score can be decomposed into similarity with each descriptor, more akin to Chen et al. (2019); Vinyals et al. (2017).”
The Secrets of Machine Learning: Ten Things You Wish You Had Known Earlier to be More Effective at Data Analysis (2019, methodology): “as computer vision or speech. There have been some eorts to make neural network models interpretable. If the neural network is interpretable, we do not need a posthoc explanation of what it is doing [11,34]. We will give two examples below. It is important to note that there is not a single denition”