CoChat

What Do You See?: Evaluation of Explainable Artificial Intelligence (XAI) Interpretability through Neural Backdoors

Open access
Lin, Yi-Shan, Lee, Wen-Chuan, Celik, Z. Berkay · 2021 · Proceedings of the 27th ACM SIGKDD Conference on Knowledge Discovery & Data Mining

The DOI matches its Crossref record; 49 of the 49 references it cites were found in a registry or an index.

Abstract

EXplainable AI (XAI) methods have been proposed to interpret how a deep neural network predicts inputs through model saliency explanations that highlight the input parts deemed important to arrive at a decision for a specific target. However, it remains challenging to quantify the correctness of their interpretability as current evaluation approaches either require subjective input from humans or incur high computation cost with automated evaluation. In this paper, we propose backdoor trigger patterns--hidden malicious functionalities that cause misclassification--to automate the evaluation of saliency explanations. Our key observation is that triggers provide ground truth for inputs to evaluate whether the regions identified by an XAI method are truly relevant to its output. Since backdoor triggers are the most important features that cause deliberate misclassification, a robust XAI method should reveal their presence at inference time. We introduce three complementary metrics for the systematic evaluation of explanations that an XAI method generates. We evaluate seven state-of-the-art model-free and model-specific post-hoc methods through 36 models trojaned with specifically crafted triggers using color, shape, texture, location, and size. We found six methods that use local explanation and feature relevance fail to completely highlight trigger regions, and only a model-free approach can uncover the entire trigger region. We made our code available at https://github.com/yslin013/evalxai.

Overview

The claims and statistics of this paper haven’t been checked yet.

Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.

A few free checks a day · papers already verified open instantly · results are public

Integrity signals

Registry

Crossref · 10.1145/3447548.3467213

Cites

49 of 49 references verified0 citation errors · 0 not checkable

Statistics

Figures

Figure 1 A trojaned model classifies any input image to 'tench' when a yellow square is attached to the image: (a) shows the saliency map that successfully highlights the informative features of the harvester, and (b) shows the saliency map that fails to highlight the trigger.
Figure 1 A trojaned model classifies any input image to 'tench' when a yellow square is attached to the image: (a) shows the saliency map that successfully highlights the informative features of the harvester, and (b) shows the saliency map that fails to highlight the trigger.CC-BY · From the PDF
Figure 2 The architecture of our XAI evaluation framework.
Figure 2 The architecture of our XAI evaluation framework.CC-BY · From the PDF
Figure 3 Illustration of saliency maps, true and detected trigger area generated by seven XAI methods.
Figure 3 Illustration of saliency maps, true and detected trigger area generated by seven XAI methods.CC-BY · From the PDF
Figure 4 Trojan triggers for multiple target attacks
Figure 4 Trojan triggers for multiple target attacksCC-BY · From the PDF
target label: 0
target label: 0CC-BY · From the PDF
true label: 727
true label: 727CC-BY · From the PDF
Partially recovered Images
Partially recovered ImagesCC-BY · From the PDF
predicted : 727 (diff: 0.0238)
predicted : 727 (diff: 0.0238)CC-BY · From the PDF
Figure 6 The recovering difference ( RD ) metric for different XAI methods applied to three neural network architectures (VGG16, ResNet50, and AlexNet): (a)-(c) are single target trojaned models, and (d)-(f) are multiple target trojaned models. RD scores increase with trigger size, indicating that XAI methods are not effective in detecting large triggers.
Figure 6 The recovering difference ( RD ) metric for different XAI methods applied to three neural network architectures (VGG16, ResNet50, and AlexNet): (a)-(c) are single target trojaned models, and (d)-(f) are multiple target trojaned models. RD scores increase with trigger size, indicating that XAI methods are not effective in detecting large triggers.CC-BY · From the PDF
Figure 7 An illustration of Backpropagation ( BP ) for trigger detection that fails to thoroughly highlight the entire trigger region when the trigger size increases. The trigger still persists and causes misclassification though we attempt to recover the detected trigger region fully.
Figure 7 An illustration of Backpropagation ( BP ) for trigger detection that fails to thoroughly highlight the entire trigger region when the trigger size increases. The trigger still persists and causes misclassification though we attempt to recover the detected trigger region fully.CC-BY · From the PDF
Figure 8 The forward-based methods ( OCC , FA and LIME ) incur much more computation cost than those of backwardbased methods ( BP , GBP , GCAM , and GGCAM).
Figure 8 The forward-based methods ( OCC , FA and LIME ) incur much more computation cost than those of backwardbased methods ( BP , GBP , GCAM , and GGCAM).CC-BY · From the PDF

Verification

Claims the paper makes

The claims and statistics of this paper haven’t been checked yet.

Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.

A few free checks a day · papers already verified open instantly · results are public

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

The self-check needs the paper’s full text.

Citations

Cites

Show all 49 references

Research works printed but not linked: 1

  1. David Alvarez Melis and Tommi Jaakkola. 2018. Towards robust interpretability with self-explaining neural networks. In Advances in Neural Information Processing Systems. 7775--7784. David Alvarez Melis and Tommi Jaakkola. 2018. Towards robust interpretability with self-explaining neural networks. In Advances in Neural Information Processing Systems. 7775--7784.

Cited by

See all 104

Metadata

Authors
Lin, Y.-S., Lee, W.-C., & Celik, Z. B.
Affiliations
  • Purdue University West Lafayette (US)
  • Apple (United States) (US)
Venue
Proceedings of the 27th ACM SIGKDD Conference on Knowledge Discovery & Data Mining
Publisher
ACM
Published
2021-08-14
Pages
1027-1035
OpenAlex
W3170542759
Cited by
64
References
40
Citations by year
2026: 4 · 2025: 22 · 2024: 16 · 2023: 15 · 2022: 6 · 2021: 1
FWCI
4.39
Type
conference paper
Language
en
Open access
Locations
ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD) (publishedVersion, cc-by)
Topics
  • Physical Sciences › Computer Science › Artificial Intelligence › Explainable Artificial Intelligence (XAI)
  • Physical Sciences › Computer Science › Artificial Intelligence › Adversarial Robustness in Machine Learning
  • Physical Sciences › Computer Science › Software › Software Testing and Debugging Techniques
Keywords
explainable AI, saliency maps, backdoor triggers, backdoor attacks, explanation evaluation, post-hoc explanation methods, adversarial robustness, misclassification, trigger detection

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

How to cite

Cite CoChat for the checks on this page, and the paper for its own findings.

This page
CoChat. (2026). Citation check: What Do You See?: Evaluation of Explainable Artificial Intelligence (XAI) Interpretability through Neural Backdoors. https://app.cochat.ai/papers/10.1145/3447548.3467213
The paper
Lin, Y.-S., Lee, W.-C., & Celik, Z. B. (2021). What Do You See?: Evaluation of Explainable Artificial Intelligence (XAI) Interpretability through Neural Backdoors. Proceedings of the 27th ACM SIGKDD Conference on Knowledge Discovery & Data Mining. https://doi.org/10.1145/3447548.3467213

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.