Explaining Image Classifiers using Statistical Fault Localization

Daniel Kroening; Hana Chockler; Xiaowei Huang; Youcheng Sun

REVIEW 3 major objections 5 minor 36 references

Explaining Image Classifiers using Statistical Fault Localization

T0 review · 3 major / 5 minor · reviewed 2026-08-14 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read The paper claims that statistical fault localization, a debugging technique from software engineering, produces image-classifier explanations that are consistently better than those of five state-of-the-art methods.

desk verdict A genuine new application of SFL to DNN explanation, weakened by missing error bars and an unvalidated ground-truth benchmark; worth reviewing but not yet definitive. read the letter →

arxiv 1908.02374 v2 pith:Q2WJOCTG submitted 2019-08-06 cs.LG cs.AI

Youcheng Sun , Hana Chockler , Xiaowei Huang , Daniel Kroening This is my paper · ORCID

classification cs.LGcs.AI

keywords deeplearningexplainableAIstatisticalfaultlocalizationimageclassificationminimalsufficientexplanationblack-boxexplainabilitysoftwaretestingTrojantriggerdetection

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper sets out to show that statistical fault localization (SFL), the same ranking formulas used to find bugs in software, can explain individual image-classifier decisions. An explanation is defined as a minimal subset of pixels that is sufficient for the network to keep the original label once every other pixel is set to the background color. DeepCover generates a test suite of randomly masked versions of the input, labels each one as "same class" or "different class", ranks all pixels by SFL suspiciousness, and adds pixels in that order until the network agrees. In comparisons against GradCAM, LIME, SHAP, RISE, and Extremal, DeepCover's explanations are claimed to be consistently better by size, by deletion behavior, by object-localization overlap, and on a synthetic ground-truth benchmark where it reaches 76.7% accuracy versus 70.7% for the second-best method. If true, this gives practitioners a cheap, black-box route to auditing what a network actually looks at.

What carries the argument

The mechanism that carries the argument is the per-pixel SFL spectrum $\langle a_{ep}, a_{ef}, a_{np}, a_{nf}\rangle$, where $e/n$ records whether a pixel is present or masked in a mutant and $p/f$ records whether the mutant keeps or changes the original label. Four standard measures—Ochiai, Tarantula, Zoltar, and Wong-II—turn these counts into a suspiciousness score, and Algorithm 2 adaptively tunes the masking fraction $\sigma$ so the mutant suite stays balanced between passing and failing cases. This ranking feeds a greedy loop that adds pixels in descending suspiciousness until the DNN outputs the original label, and that set is returned as the explanation. Because the loop stops at the first sufficient set, the result is an approximation of a minimal explanation rather than an exact one; exact computation is DP-complete, i.e. intractable.

What would settle it

Take a Chimera image, mask exactly the embedded red-panda patch by setting all other pixels to the background color, and check whether the network still outputs "red panda"; if it does not, the patch is not sufficient, and the ground-truth assumption behind the 76.7% IoU comparison fails.

Watch

Extended reading notes

Core claim

The paper's central claim is that SFL measures transfer from program statements to input pixels without modification. For a fixed image classified as $y$, DeepCover constructs many masked mutants and records, for each pixel, how often it is present or absent in mutants that keep $y$ ("passing") and mutants that leave $y$ ("failing"). Applying measures such as Ochiai, Tarantula, Zoltar, and Wong-II turns these four counts into a suspiciousness score per pixel. Sorting by that score and greedily adding pixels until the masked image is again classified as $y$ produces an approximation of a minimal sufficient explanation. The paper reports that this explanation is better than the five comparison methods on every evaluation axis, with the Chimera benchmark giving DeepCover 76.7% accuracy at IoU ≥ 0.5, 6 percentage points above the second-best method, Extremal.

Load-bearing premise

The Chimera benchmark assumes that the superimposed red-panda patch is the exact ground truth: every pixel inside the patch is explanatory and every pixel outside it is not, so any explanation overlapping the patch is counted correct and any overlap outside is counted wrong.

Editorial extensions

If this is right

Because DeepCover only needs the network's output labels on masked images, it works for any classifier that can be queried as a black box, without gradients or internal activations.
Explanations for ImageNet-scale images can be produced in time linear in the number of mutants and pixels, making the approach practical on a laptop.
The same ranking can serve as a deletion test: masking only 2% of the highest-ranked pixels changes classification in 60% of images, far fewer pixels than the second-best method requires.
On the Trojan-trigger dataset, the top 8% of DeepCover-ranked pixels localize 80% of the planted triggers with IoU ≥ 0.5, suggesting the method can support security auditing of deployed models.
Because the explanation tracks model parameters during training, DeepCover satisfies the sanity requirement that explanations should change when the model changes, and it can reveal cases where a correct label is reached for the wrong reason.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

The paper's explanation metric rewards smallness and sufficiency, but a direct human study could test whether DeepCover's minimal explanations are more or less trusted than redundant ones; the authors cite evidence that people sometimes prefer redundancy.
The Chimera ground-truth assumption could be stress-tested by constructing synthetic images where the planted object is necessary but not sufficient for the label; if the network then relies on context, IoU against the patch would systematically understate or overstate quality.
Because the mutant generator treats any pixel mutation as a program statement, the same pipeline should transfer to text and tabular data by replacing pixel masking with token or feature masking, giving a domain-generic black-box explainer.
A natural algorithmic next step is to use multiple SFL measures together or to select a measure per image, since the paper evaluates four fixed formulas but notes that hundreds exist.

Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

Colleague,

This paper applies statistical fault localization (SFL), a software debugging technique, to explaining image classifiers. That transfer is new as far as I know, and the authors back it with a working tool (DeepCover), a public benchmark, and honest discussion of limits. If you work on explainability, this is worth having on your radar.

What is good: the explanation notion is clearly defined as a minimal sufficient subset of pixels. The algorithm is simple, black-box, and linear-time. The evaluation is broad—explanation size, deletion/masking, weakly supervised localization, a synthetic Chimera benchmark, and trojan-trigger detection—and across most of these DeepCover leads. The tool and benchmark are concrete artifacts others can build on. The authors also state in Section 6.6 that the parameters (σ, ε, m) were chosen empirically and may be overfitted; I appreciate that admission.

The soft spots are real. There are no confidence intervals or significance tests anywhere, so the reported gaps—e.g., 76.7% vs 70.7% for Extremal—might shift with a different sample or random seed. The parameters are indeed tuned on the same type of data used in the evaluation, so the comparisons are not held-out. And the Chimera benchmark is the weakest link: the authors assume the embedded red panda patch is exactly the ground-truth explanation, but they never test whether the patch alone (with other pixels set to background) is sufficient under their own Definition 1, nor whether the host image contributes. If either fails, the IoU scores in Table 1 are not a faithful measure, and the 6% advantage over Extremal is on shaky ground. Interestingly, RISE outperforms DeepCover at IoU ≥ 0.7 in that same table, which complicates the "consistently better" claim.

None of this sinks the paper. The size and masking experiments, and the trojan result, stand somewhat independently of the Chimera assumption. But the empirical case is not as strong as the abstract suggests. I would send this to a serious referee, expecting heavy revision on the evaluation: add error bars, validate or fix the parameter choices, and either strengthen the Chimera ground-truth or soften the claim.

For a reading group, it would generate good discussion about what counts as ground truth in explanation evaluation. I would cite it in my own work as a representative black-box method.

Best.

Referee Report

3 major / 5 minor

Summary. The paper proposes DeepCover, a black-box method for explaining image classifier decisions, built on statistical fault localization (SFL) from software engineering. It defines an explanation as a minimal subset of pixels sufficient for the same classification, generates a test suite of randomly mutated images, ranks pixels by four SFL measures (Ochiai, Tarantula, Zoltar, Wong-II), and greedily grows a sufficient pixel set. The method is evaluated against GradCAM, LIME, SHAP, RISE, and Extremal using explanation size, deletion-based sensitivity, weakly supervised object localization, a synthetic Chimera benchmark with known ground truth, and Trojan trigger localization. The headline claim is that DeepCover achieves 76.7% accuracy on the Chimera benchmark, 6% better than the second-best method.

Significance. The paper is a novel and potentially impactful cross-pollination of statistical fault localization and explainable AI. DeepCover is black-box, linear-time in the image and test-suite size, and the authors provide an open-source implementation. The construction of a synthetic Chimera benchmark with known ground truth and the successful localization of Trojan triggers are valuable contributions that go beyond pixel-ranking comparisons. If the empirical claims hold, the method offers a competitive and efficient alternative to existing explanation tools. The main weaknesses are lack of statistical rigor in the evaluation and an unvalidated ground-truth assumption in the Chimera benchmark, both of which are fixable.

major comments (3)

[Section 6.4, Table 1] The Chimera benchmark assumes that the embedded 'red panda' patch is the exact ground truth explanation. Under the paper's own Definition 1, this requires that the patch alone (with all other pixels set to background) is sufficient for the red-panda classification, and that no pixels outside the patch contribute to the classification. The paper only verifies that the composed image retains the red-panda label with MobileNet and VGG16. If the network uses background cues from the host image, DeepCover's top-ranked pixels outside the patch are legitimate explanations and the IoU metric unfairly penalizes them; conversely, if the patch contains irrelevant pixels, the metric rewards including them. This assumption is load-bearing for the headline 76.7% vs. 70.7% result, so the authors should test the patch's sufficiency and analyze whether top-ranked pixels outside the patch are truly non-explanatory.
[Section 6.3, Figures 3-5, Table 1] The empirical evaluation reports no confidence intervals, significance tests, or variance across repeated runs. This matters because Algorithm 2 is randomized (the initial σ is sampled, and pixel masks are random), and the parameters σ=1/5, ε=1/6, and m=2000 are chosen empirically with no held-out validation, as acknowledged in Section 6.6. The reported advantages of DeepCover over the second-best methods in Figures 3-5 and Table 1 could therefore be within noise. Please provide repeated-run statistics (e.g., means and standard deviations over several independent runs with different random seeds) and a significance test, or otherwise demonstrate that the differences are stable.
[Section 5.2 and Figure 3] Algorithm 1 constructs an explanation by adding pixels until the classification matches, but the paper acknowledges that the result is not necessarily minimal under Definition 1. This non-minimality is not quantified, yet Figure 3 compares the sizes of DeepCover's explanations with those of other methods. The comparison therefore conflates the quality of the ranking with the greedy construction rule: a different stopping criterion (e.g., attempting to remove redundant pixels after the greedy phase) could change the reported sizes. Please report the amount of redundancy in P_exp (for example, the fraction of pixels that can be removed while preserving the classification) and discuss how this affects the size-based comparison.

minor comments (5)

[Throughout] Tool names should be capitalized consistently (e.g., 'GradCAM', 'LIME', 'SHAP', 'RISE', 'Extremal'), including in the Abstract and captions.
[Section 6.1, Figure 6] The test-suite size is denoted m in Algorithm 2 and Section 6.1, but Figure 6 labels it 'n'. Please unify the notation.
[Section 6.4, Table 1 caption] The text says 'intersection of union'; the standard term is 'intersection over union' (IoU). Please correct.
[Algorithm 2] Line 2 says 'σ← sample in the range (0, 1)' but does not specify the distribution. Please state whether the sample is uniform and how the random seed is handled.
[Equation (2c)] The Zoltar formula as typeset is ambiguous in the denominator. Add explicit parentheses to clarify the precedence of the terms.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: DeepCover's SFL ranking and explanation construction are self-contained, and the Chimera and Trojan benchmarks provide external, independently planted ground truths.

full rationale

The paper's derivation chain is self-contained and does not reduce to its own inputs. DeepCover defines an explanation as a minimal subset of pixels sufficient for the same classification (Definition 1), ranks pixels using standard external SFL formulas (Ochiai, Tarantula, Zoltar, Wong-II in Eq. 2), and constructs an explanation by adding top-ranked pixels until the DNN returns the original label (Algorithm 1). Sufficiency is enforced by the stopping condition, and minimality is only approximated, as the paper explicitly acknowledges in Section 5.2. The evaluation compares against external tools using independent metrics: explanation size, addition/deletion curves, weakly supervised object localization IoU, the Chimera benchmark with a red panda patch planted by the authors, and the Trojan trigger from prior external work [17]. The Chimera ground truth rests on an assumption that the embedded patch is the true explanation, but this is an unverified empirical assumption about the benchmark, not a circular reduction: the patch is constructed independently of DeepCover's ranking, and the IoU comparison is computed against that externally planted region. The self-citations in the paper ([13], [15], [16]) support background claims about SFL measures and single-bug optimality; they are not used to justify DeepCover's central output or to forbid alternatives. No equation or fitted parameter is renamed as a prediction, and no load-bearing claim is justified solely by a self-citation chain. Accordingly, the paper merits a circularity score of 0.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

The method's success depends on empirically chosen test-generation parameters (sigma=1/5, epsilon=1/6, m=2000), a greedy minimality approximation, and benchmark labeling assumptions. None of these are derived from first principles or verified on independent data. The black-box assumption is mild and standard for explainability. No new physical entities are introduced.

free parameters (3)

sigma (initial mask fraction) = 1/5
Chosen empirically (Section 6.1); affects the shape of the mutant distribution and thus the SFL scores.
epsilon (mask fraction step) = 1/6
Chosen empirically (Section 6.1); controls how quickly the adaptive test generation shifts between passing and failing mutants.
m (test suite size) = 2000
Chosen empirically (Section 6.1); larger m improves stability but costs more forward passes.

assumptions (4)

domain assumption The DNN can be treated as a black box accessible only through input-output queries.
The algorithm in Section 3.1 is designed to be independent of internal structure; ranking relies only on classification agreement of mutated inputs.
ad hoc to paper The greedy addition of pixels in descending SFL rank order approximates a minimal sufficient explanation.
Section 5.2 states the result may not be minimal and argues redundancy is likely small, but no bound or empirical quantification is given.
ad hoc to paper Balancing passing and failing mutants in the test suite yields informative SFL spectra.
Algorithm 2 adaptively adjusts the mask fraction to balance labels, based on a debugging rule of thumb, with no theoretical justification (Section 5.1).
ad hoc to paper The Chimera patch annotations are exact ground truth for explanations.
Section 6.4 assumes that the superimposed red panda region contains the explanatory pixels and all other pixels do not explain the classification.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Explaining Image Classifiers using Statistical Fault Localization." pith.science (2026). https://pith.science/paper/Q2WJOCTG

@misc{pith2026190802374,
  author       = {Pith},
  title        = {Pith review of: Explaining Image Classifiers using Statistical Fault Localization},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/Q2WJOCTG}},
  note         = {Machine review of arXiv:1908.02374}
}

read the original abstract

The black-box nature of deep neural networks (DNNs) makes it impossible to understand why a particular output is produced, creating demand for "Explainable AI". In this paper, we show that statistical fault localization (SFL) techniques from software engineering deliver high quality explanations of the outputs of DNNs, where we define an explanation as a minimal subset of features sufficient for making the same decision as for the original input. We present an algorithm and a tool called DeepCover, which synthesizes a ranking of the features of the inputs using SFL and constructs explanations for the decisions of the DNN based on this ranking. We compare explanations produced by DeepCover with those of the state-of-the-art tools GradCAM, LIME, SHAP, RISE and Extremal and show that explanations generated by DeepCover are consistently better across a broad set of experiments. On a benchmark set with known ground truth, DeepCover achieves 76.7% accuracy, which is 6% better than the second best Extremal.

Figures

Figures reproduced from arXiv: 1908.02374 by the authors.

**Figure 2.** Explanations of the DNN at different training stages: the 1st column are the original images and the subsequent columns give the explanations for a particular training iteration (CIFAR10 validation data set) the progress of the training reflected in the explanations of DNN’s classification of the first image as a ‘cat’). This result reflects that the DNN is being trained to learn features of different classes of … view at source ↗

**Figure 4.** Misclassification vs percentage of masked pixels for different tools [PITH_FULL_IMAGE:figures/full_fig_p011_4.png] view at source ↗

**Figure 6.** Explanations for the ‘Welsh springer spaniel’ by DeepCover and rise with varying number of samples (i.e. n) Next, we present a synthetic benchmark (Sec. 6.4) and a security application (Sec. 6.5). 6.4 Generating “ground truth” with a Chimera benchmark The biggest challenge in evaluating explanations for DNNs (and even for human decision making) is the lack of the ground truth. Human evaluations of the explanations r… view at source ↗

Figures from the paper (2 more)

**Figure 7.** Figure 7: Examples of embedding the red panda IoU≥0.5 IoU≥0.6 IoU≥0.7 DeepCover 76.7% 54.9% 9.8% extremal 70.7% 21.5% 2.2% rise 55.8% 42.9% 25.7% gradcam 0% 0% 0% [PITH_FULL_IMAGE:figures/full_fig_p013_7.png]

**Figure 8.** Figure 8: Applying DeepCover to Trojaning attacks on VGG Face. The Trojan trigger is the square shape in the lower right corner of the image; the DeepCover explanation for the Trojan behaviour is on the right. as many as 80% of the (ground truth) Trojan triggers are successively…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

36 extracted references · 34 canonical work pages

[1]

In: Advances in Neural Information Processing Systems

Adebayo, J., Gilmer, J., Muelly, M., Goodfellow, I., Hardt, M., Kim, B.: Sanity checks for saliency maps. In: Advances in Neural Information Processing Systems. pp. 9505–9515 (2018)

2018
[2]

In: Uncertainty in Artiﬁcial Intelligence (UAI)

Chajewska, U., Halpern, J.Y.: Deﬁning explanation in probabilistic systems. In: Uncertainty in Artiﬁcial Intelligence (UAI). pp. 62–71. Morgan Kaufmann (1997)

work page 1997
[3]

In: International Conference on Machine Learning (ICML)

Chen, J., Song, L., Wainwright, M., Jordan, M.: Learning to explain: An information- theoretic perspective on model interpretation. In: International Conference on Machine Learning (ICML). vol. 80, pp. 882–891. PMLR (2018)

work page 2018
[4]

In: Security and Privacy (S&P)

Datta, A., Sen, S., Zick, Y.: Algorithmic transparency via quantitative input inﬂuence: Theory and experiments with learning systems. In: Security and Privacy (S&P). pp. 598–617. IEEE (2016)

work page 2016
[5]

Eiter, T., Lukasiewicz, T.: Complexity results for explanations in the structural- model approach. Artif. Intell. 154(1-2), 145–198 (2004)

work page 2004
[6]

In: International Conference on Computer Vision (ICCV)

Fong, R., Patrick, M., Vedaldi, A.: Understanding deep networks via extremal perturbations and smooth masks. In: International Conference on Computer Vision (ICCV). pp. 2950–2958. IEEE (2019)

work page 2019
[7]

MIT Press (1988)

G¨ ardenfors, P.: Knowledge in Flux. MIT Press (1988)

work page 1988
[8]

Gonzalez-Sanchez, A.: Automatic error detection techniques based on dynamic invariants. M.S. Thesis, Delft University of Technology, The Netherlands (2007)

work page 2007

Show all 36 references

[9]

https://www.darpa.mil/program/explainable-artiﬁcial-intelligence (2017), Defense Advanced Research Projects Agency

Gunning, D.: Explainable artiﬁcial intelligence (XAI) – program information. https://www.darpa.mil/program/explainable-artiﬁcial-intelligence (2017), Defense Advanced Research Projects Agency

2017
[10]

Part I: Causes

Halpern, J.Y., Pearl, J.: Causes and explanations: a structural-model approach. Part I: Causes. British Journal for the Philosophy of Science 56(4) (2005)

2005
[11]

Part II: Explanations

Halpern, J.Y., Pearl, J.: Causes and explanations: a structural-model approach. Part II: Explanations. British Journal for the Philosophy of Science 56(4) (2005)

2005
[12]

Free Press (1965)

Hempel, C.G.: Aspects of Scientiﬁc Explanation. Free Press (1965)

1965
[13]

Computer Science Review 37, 100270 (2020)

Huang, X., Kroening, D., Ruan, W., Sharp, J., Sun, Y., Thamo, E., Wu, M., Yi, X.: A survey of safety and trustworthiness of deep neural networks: Veriﬁcation, testing, adversarial attack and defence, and interpretability. Computer Science Review 37, 100270 (2020)

2020
[14]

In: Proceedings of ASE

Jones, J.A., Harrold, M.J.: Empirical evaluation of the Tarantula automatic fault- localization technique. In: Proceedings of ASE. pp. 273–282. ACM (2005)

2005
[15]

In: Fundamental Approaches to Software Engineering (FASE)

Landsberg, D., Chockler, H., Kroening, D., Lewis, M.: Evaluation of measures for statistical fault localisation and an optimising scheme. In: Fundamental Approaches to Software Engineering (FASE). LNCS, vol. 9033, pp. 115–129. Springer (2015)

2015
[16]

In: Fundamental Approaches to Software Engineering (FASE)

Landsberg, D., Sun, Y., Kroening, D.: Optimising spectrum based fault localisation for single fault programs using speciﬁcations. In: Fundamental Approaches to Software Engineering (FASE). LNCS, vol. 10802, pp. 246–263. Springer (2018)

2018
[17]

In: Network and Distributed System Security Symposium (NDSS)

Liu, Y., Ma, S., Aafer, Y., Lee, W., Zhai, J., Wang, W., Zhang, X.: Trojaning attack on neural networks. In: Network and Distributed System Security Symposium (NDSS). The Internet Society (2018)

2018
[18]

In: Advances in Neural Information Processing Systems

Lundberg, S.M., Lee, S.I.: A uniﬁed approach to interpreting model predictions. In: Advances in Neural Information Processing Systems. pp. 4765–4774 (2017)

2017
[19]

ACM TOSEM 20(3), 11 (2011)

Naish, L., Lee, H.J., Ramamohanarao, K.: A model for spectra-based software diagnosis. ACM TOSEM 20(3), 11 (2011)

2011
[20]

In: Proceedings of the International Conference on Software Engineering (ICSE) (2020) 16 Youcheng Sun, Hana Chockler, Xiaowei Huang, and Daniel Kroening

Noller, Y., P˘ as˘ areanu, C.S., B¨ ohme, M., Sun, Y., Nguyen, H.L., Grunske, L.: HyDiﬀ: Hybrid diﬀerential software analysis. In: Proceedings of the International Conference on Software Engineering (ICSE) (2020) 16 Youcheng Sun, Hana Chockler, Xiaowei Huang, and Daniel Kroening

2020
[21]

Bulletin of Japanese Society of Scientiﬁc Fisheries 22, 526– 530 (1957)

Ochiai, A.: Zoogeographic studies on the soleoid ﬁshes found in Japan and its neighbouring regions. Bulletin of Japanese Society of Scientiﬁc Fisheries 22, 526– 530 (1957)

1957
[22]

In: International Conference on Machine Learning

Odena, A., Olsson, C., Andersen, D., Goodfellow, I.: Tensorfuzz: Debugging neural networks with coverage-guided fuzzing. In: International Conference on Machine Learning. pp. 4901–4911 (2019)

2019
[23]

Distill (2018)

Olah, C., Satyanarayan, A., Johnson, I., Carter, S., Schubert, L., Ye, K., Mordvint- sev, A.: The building blocks of interpretability. Distill (2018)

2018
[24]

Morgan Kaufmann (1988)

Pearl, J.: Probabilistic Reasoning in Intelligent Systems. Morgan Kaufmann (1988)

1988
[25]

In: British Machine Vision Conference (BMVC)

Petsiuk, V., Das, A., Saenko, K.: RISE: randomized input sampling for explanation of black-box models. In: British Machine Vision Conference (BMVC). BMVA Press (2018)

2018
[26]

Nature 568(7753), 477 (2019)

Rahwan, I., Cebrian, M., Obradovich, N., Bongard, J., Bonnefon, J.F., Breazeal, C., Crandall, J.W., Christakis, N.A., Couzin, I.D., Jackson, M.O., et al.: Machine behaviour. Nature 568(7753), 477 (2019)

2019
[27]

Why should I trust you?

Ribeiro, M.T., Singh, S., Guestrin, C.: “Why should I trust you?” Explaining the predictions of any classiﬁer. In: Knowledge Discovery and Data Mining (KDD). pp. 1135–1144. ACM (2016)

2016
[28]

University of Minnesota Press (1989)

Salmon, W.C.: Four Decades of Scientiﬁc Explanation. University of Minnesota Press (1989)

1989
[29]

In: International Conference on Computer Vision (ICCV)

Selvaraju, R.R., Cogswell, M., Das, A., Vedantam, R., Parikh, D., Batra, D.: Grad- CAM: Visual explanations from deep networks via gradient-based localization. In: International Conference on Computer Vision (ICCV). pp. 618–626. IEEE (2017)

2017
[30]

In: International Conference on Machine Learning (ICML)

Shrikumar, A., Greenside, P., Kundaje, A.: Learning important features through propagating activation diﬀerences. In: International Conference on Machine Learning (ICML). vol. 70, pp. 3145–3153. PMLR (2017)

2017
[31]

In: Proceedings of the 33rd ACM/IEEE International Conference on Automated Software Engineering, ASE

Sun, Y., Wu, M., Ruan, W., Huang, X., Kwiatkowska, M., Kroening, D.: Con- colic testing for deep neural networks. In: Proceedings of the 33rd ACM/IEEE International Conference on Automated Software Engineering, ASE. pp. 109–119 (2018)

2018
[32]

In: International Conference on Robotics and Automation (ICRA)

Sun, Y., Zhou, Y., Maskell, S., Sharp, J., Huang, X.: Reliability validation of learning enabled vehicle tracking. In: International Conference on Robotics and Automation (ICRA). IEEE (2020)

2020
[33]

IEEE TSE 42(8), 707–740 (2016)

Wong, W.E., Gao, R., Li, Y., Abreu, R., Wotawa, F.: A survey on software fault localization. IEEE TSE 42(8), 707–740 (2016)

2016
[34]

In: Computer Software and Applications Conference (COMPSAC)

Wong, W.E., Qi, Y., Zhao, L., Cai, K.: Eﬀective fault localization using code coverage. In: Computer Software and Applications Conference (COMPSAC). pp. 449–456 (2007)

2007
[35]

Psychonomic Bulletin & Review 24(5), 1488–1500 (2017)

Zemla, J.C., Sloman, S., Bechlivanidis, C., Lagnado, D.A.: Evaluating everyday explanations. Psychonomic Bulletin & Review 24(5), 1488–1500 (2017)

2017
[36]

The Verge (2016), https://www.theverge.com/2016/2/29/11134344/ google-self-driving-car-crash-report

Ziegler, C.: A Google self-driving car caused a crash for the ﬁrst time. The Verge (2016), https://www.theverge.com/2016/2/29/11134344/ google-self-driving-car-crash-report

2016

Pith tools

Reviewed August 14, 2026 · model on record in the stance chip above.

[1] [1]

In: Advances in Neural Information Processing Systems

Adebayo, J., Gilmer, J., Muelly, M., Goodfellow, I., Hardt, M., Kim, B.: Sanity checks for saliency maps. In: Advances in Neural Information Processing Systems. pp. 9505–9515 (2018)

2018

[2] [2]

In: Uncertainty in Artiﬁcial Intelligence (UAI)

Chajewska, U., Halpern, J.Y.: Deﬁning explanation in probabilistic systems. In: Uncertainty in Artiﬁcial Intelligence (UAI). pp. 62–71. Morgan Kaufmann (1997)

work page 1997

[3] [3]

In: International Conference on Machine Learning (ICML)

Chen, J., Song, L., Wainwright, M., Jordan, M.: Learning to explain: An information- theoretic perspective on model interpretation. In: International Conference on Machine Learning (ICML). vol. 80, pp. 882–891. PMLR (2018)

work page 2018

[4] [4]

In: Security and Privacy (S&P)

Datta, A., Sen, S., Zick, Y.: Algorithmic transparency via quantitative input inﬂuence: Theory and experiments with learning systems. In: Security and Privacy (S&P). pp. 598–617. IEEE (2016)

work page 2016

[5] [5]

Eiter, T., Lukasiewicz, T.: Complexity results for explanations in the structural- model approach. Artif. Intell. 154(1-2), 145–198 (2004)

work page 2004

[6] [6]

In: International Conference on Computer Vision (ICCV)

Fong, R., Patrick, M., Vedaldi, A.: Understanding deep networks via extremal perturbations and smooth masks. In: International Conference on Computer Vision (ICCV). pp. 2950–2958. IEEE (2019)

work page 2019

[7] [7]

MIT Press (1988)

G¨ ardenfors, P.: Knowledge in Flux. MIT Press (1988)

work page 1988

[8] [8]

Gonzalez-Sanchez, A.: Automatic error detection techniques based on dynamic invariants. M.S. Thesis, Delft University of Technology, The Netherlands (2007)

work page 2007