Deformable ProtoPNet: An Interpretable Image Classifier Using Deformable Prototypes

Donnelly, Jon; Barnett, Adam S.; Chen, Chaofan

doi:10.48550/arxiv.2111.15000

Cited by 4 publications

(7 citation statements)

References 25 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Figure 7 illustrates the general pipeline for deriving a class prediction from the similarity score between different parts of the input image and a set of learned prototypes. Based on ProtoPNet, Donnelly et al [28] introduced the Deformable ProtoPNet. This prototypical case-based interpretable neural network provided spatially flexible deformable prototypes, i.e., prototypes that can change their relative position to detect semantically similar parts of an input image.…”

Section: Propotypesmentioning

confidence: 99%

Explainable Deep Learning Methods in Medical Imaging Diagnosis: A Survey

Patrício¹,

Neves²,

Teixeira³

2022

Preprint

View full text Add to dashboard Cite

The remarkable success of deep learning has prompted interest in its application to medical diagnosis. Even tough state-ofthe-art deep learning models have achieved human-level accuracy on the classification of different types of medical data, these models are hardly adopted in clinical workflows, mainly due to their lack of interpretability. The black-box-ness of deep learning models has raised the need for devising strategies to explain the decision process of these models, leading to the creation of the topic of eXplainable Artificial Intelligence (XAI). In this context, we provide a thorough survey of XAI applied to medical diagnosis, including visual, textual, and example-based explanation methods. Moreover, this work reviews the existing medical imaging datasets and the existing metrics for evaluating the quality of the explanations . Complementary to most existing surveys, we include a performance comparison among a set of report generation-based methods. Finally, the major challenges in applying XAI to medical imaging are also discussed.CCS Concepts: • Applied computing → Health care information systems.

show abstract

Section: Propotypesmentioning

confidence: 99%

Explainable Deep Learning Methods in Medical Imaging Diagnosis: A Survey

Patrício¹,

Neves²,

Teixeira³

2022

Preprint

View full text Add to dashboard Cite

show abstract

“…Recent work found that re-ranking kNN's shortlisted candidates using the patch-wise similarity between the query and training set examples can further improve classification accuracy on OOD data for some image matching tasks [29,79,75] such as face identification [29]. Furthermore, patch-level comparison is also useful in prototype-based bird classifiers [17,22]. Inspired by these prior successes and the fact that EMD-Corr and CHM-Corr base the patch-wise similarity of two images on only 5 pairs of patches instead of all 49×49 = 2,401 pairs as in [29,79,75], here we test whether our two proposed re-rankers are able to improve the test-set accuracy and robustness over kNN.…”

Section: Visual Correspondence-based Explanations Improve Knn Robustn...mentioning

confidence: 99%

“…Our EMD-Corr and CHM-Corr present a novel combination of heatmap-based and prototype-based XAI approaches. None of the prior prototype-based XAI methods that operate at the patch level [17,53,79,22] (see Table 3 in [22]) were tested on humans yet. Also, in preliminary tests, we find their explanation formats too dense (i.e., showing over 10 prototypes [17], 9 correspondence pairs per image [22], or an entire prototype tree to humans [52,53]) to be useful for lay users.…”

Section: Related Workmentioning

confidence: 99%

Visual correspondence-based explanations improve AI robustness and human-AI team accuracy

Nguyen¹,

Taesiri²,

Nguyen³

2022

Preprint

View full text Add to dashboard Cite

Explaining artificial intelligence (AI) predictions is increasingly important and even imperative in many high-stakes applications where humans are the ultimate decision makers. In this work, we propose two novel architectures of self-interpretable image classifiers that first explain, and then predict (as opposed to post-hoc explanations) by harnessing the visual correspondences between a query image and exemplars. Our models consistently improve (by 1 to 4 points) on out-of-distribution (OOD) datasets while performing marginally worse (by 1 to 2 points) on in-distribution tests than ResNet-50 and a k-nearest neighbor classifier (kNN). Via a large-scale, human study on ImageNet and CUB, our correspondence-based explanations are found to be more useful to users than kNN explanations. Our explanations help users more accurately reject AI's wrong decisions than all other tested methods. Interestingly, for the first time, we show that it is possible to achieve complementary human-AI team accuracy (i.e., that is higher than either AI-alone or human-alone), in ImageNet and CUB image classification tasks. * Equal contribution. Listing order is random. GN led the development of EMD-Corr and human studies on Gorilla. MRT led the development of CHM-Corr, pilot studies on HuggingFace, and the analysis of human-study data from Gorilla. AN advised the project. All authors contributed to data analysis and paper writing.Preprint. Under review.

show abstract

“…Additionally, we further design a measurement to quantitatively evaluate the visual-based interpretation. [35,13] extend ProtoPNet in various directions. For instance, ProtoTree [27] combines prototype learning with decision trees, resulting in an interpretable decision path consisting of prototypes.…”

Section: Introductionmentioning

confidence: 99%

“…We notice two new variants of ProtoPNet, ProtoPool[35] and Deformable ProtoPNet[13]. We do not include these two methods for comparison as their papers are not formally published and their codes are still unavailable.…”

mentioning

confidence: 99%

Learnable Visual Words for Interpretable Image Recognition

Xiao¹,

Ding²,

Liu³

2022

Preprint

View full text Add to dashboard Cite

To interpret deep models' predictions, attention-based visual cues are widely used in addressing why deep models make such predictions. Beyond that, the current research community becomes more interested in reasoning how deep models make predictions, where some prototype-based methods employ interpretable representations with their corresponding visual cues to reveal the black-box mechanism of deep model behaviors. However, these pioneering attempts only either learn the category-specific prototypes and deteriorate their generalizing capacities, or demonstrate several illustrative examples without a quantitative evaluation of visual-based interpretability with further limitations on their practical usages. In this paper, we revisit the concept of visual words and propose the Learnable Visual Words (LVW) to interpret the model prediction behaviors with two novel modules: semantic visual words learning and dual fidelity preservation. The semantic visual words learning relaxes the category-specific constraint, enabling the general visual words shared across different categories. Beyond employing the visual words for prediction to align visual words with the base model, our dual fidelity preservation also includes the attention guided semantic alignment that encourages the learned visual words to focus on the same conceptual regions for prediction. Experiments on six visual benchmarks demonstrate the superior effectiveness of our proposed LVW in both accuracy and model interpretation over the state-of-the-art methods. Moreover, we elaborate on various in-depth analyses to further explore the learned visual words and the generalizability of our method for unseen categories.

show abstract

Deformable ProtoPNet: An Interpretable Image Classifier Using Deformable Prototypes

Cited by 4 publications

References 25 publications

Explainable Deep Learning Methods in Medical Imaging Diagnosis: A Survey

Explainable Deep Learning Methods in Medical Imaging Diagnosis: A Survey

Visual correspondence-based explanations improve AI robustness and human-AI team accuracy

Learnable Visual Words for Interpretable Image Recognition

Contact Info

Product

Resources

About