Pith. sign in

REVIEW 3 major objections 5 minor 28 references

Feature-Guided Neighbor Selection for Non-Expert Evaluation of Model Predictions

T0 review · 3 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read Selecting class-prototype-aligned explanation examples helps non-experts spot AI model errors.

desk verdict FGNS is a plausible, modestly supported method for helping non-experts spot misclassifications, but the paper's quantitative validation is partly circular and the judgment-accuracy analysis has an unspecified Likert-to-binary threshold. read the letter →

arxiv 2507.06029 v2 pith:6BUFBMW5 submitted 2025-07-08 cs.AI

Courtney Ford , Mark T. Keane This is my paper · ORCID
classification cs.AI
keywords ExplainableAIexample-basedexplanationsfeatureimportancek-nearestneighborsclassprototypesnon-expertusersKannada-MNISTmodelevaluation
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

This paper introduces Feature-Guided Neighbor Selection (FGNS), a post hoc explanation method that decides which example images to show beside a model prediction. Instead of choosing the nearest neighbors in feature space, FGNS re-ranks candidates by how far they deviate from a prototype of the predicted class, weighting the comparison with features that LIME highlights locally and SAGE validates globally. In a study with 98 non-experts judging Kannada digit predictions, participants given FGNS examples rated misclassifications as less correct and made faster, more accurate right/wrong judgments than participants given ordinary k-NN examples, while rating correct predictions almost identically. The paper claims that class-representative, prototype-aligned neighbors support people without domain expertise in validating AI decisions, and that this performance gain does not automatically raise perceived trust.

What carries the argument

The central object is the feature-guided ranking loss $L_{\text{feature}} = \rho \sum_{i=1}^k \| M_i \odot (I_{\text{candidate}} - P_c) \|_2^2$, where $P_c$ is the pixel-wise median prototype of the predicted class $c$, $M_i$ are binary feature masks obtained by aggregating LIME's local explanations and validating them with SAGE's global importance, and $\odot$ is element-wise multiplication. This loss measures how much a candidate training image deviates from the class prototype in the regions the model treats as characteristic, and FGNS selects the three lowest-loss neighbors as the explanation. The six-step pipeline (local LIME extraction, class-level aggregation, SAGE validation, clustering and diversity filtering, prototype construction, and re-ranked selection) turns the abstract idea of class representativeness into a concrete, computable neighbor ordering.

What would settle it

Re-analyse or replicate the user study with an explicit, pre-registered binary threshold (for example, ratings 1-2 count as 'incorrect' and 3-5 as 'correct') and check whether FGNS still beats k-NN on judgment accuracy and response time; alternatively, run the same paradigm on natural images with less well-defined class prototypes and see whether the error-detection advantage disappears.

Watch

Extended reading notes

Core claim

The central claim is that FGNS improves non-expert evaluation of model predictions by selecting neighbors that reflect class characteristics rather than merely minimizing feature-space distance. FGNS constructs a class prototype from the pixel-wise median of training images and integrates LIME's local superpixel attributions with SAGE's global importance scores into binary feature masks. It then ranks training instances for the predicted class by the L2 distance from each candidate to the prototype, weighted by those masks, and displays the three lowest-loss examples. In the user study this produced a significant Explanation Type $\times$ Classification Type interaction on correctness ratings, $F(1,98)=9.11$, $p=.003$, with FGNS users rating misclassifications as less correct ($M=2.49$ vs $2.83$) while rating correct classifications nearly identically ($M=4.33$ vs $4.35$); it also yielded better true/false judgment rates ($0.70$ vs $0.65$ true, $0.17$ vs $0.21$ false) and faster responses ($12.93$s vs $15.68$s per item). Quantitatively, FGNS neighbors sat closer to class prototypes and clustered more tightly than k-NN neighbors, supporting the interpretation that prototype alignment, not proximity, drives the user-level effect.

Load-bearing premise

The headline user-study results depend on an unstated rule for converting 5-point correctness ratings into binary right/wrong judgments, so the reported accuracy gains over k-NN hinge on a threshold the paper never specifies.

Editorial extensions

If this is right

  • For image classification tasks where users lack domain expertise, explanation examples should be selected for prototype alignment rather than feature-space proximity.
  • FGNS-style neighbor selection can replace proximity-based k-NN ranking in twin-system XAI architectures without retraining the predictor.
  • Prototype-guided explanations improve error detection without inflating trust or satisfaction, potentially reducing overreliance while keeping users appropriately skeptical.
  • FGNS neighbor sets are more consistent across queries (lower variance around the class prototype), making explanation output more stable and interpretable.

Reading between the lines

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

  • If prototype-aligned neighbors improve non-expert error detection on unfamiliar script, the same design principle may extend to other unfamiliar perceptual domains such as medical or satellite imagery, where class prototypes can be estimated from training data; this is an untested implication.
  • The observed dissociation between improved task performance and unchanged self-reported trust suggests trust questionnaires may be insensitive to explanation quality; a testable follow-up would measure behavioural reliance, such as how often users accept or override predictions.
  • Because FGNS depends on well-formed prototypes and clean superpixels, its advantage could shrink on natural images with high intra-class variation; a direct comparison on a more diverse dataset would map that boundary.
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.

Referee Report

3 major / 5 minor

Summary. The paper introduces Feature-Guided Neighbor Selection (FGNS), a post hoc explanation method that re-ranks k-NN candidate neighbors using a feature-mask-weighted distance to a class prototype, where the masks come from aggregating LIME explanations and validating them with SAGE global importance. The method is evaluated on Kannada-MNIST with two complementary analyses: a quantitative comparison of FGNS versus k-NN neighbor properties (distances to query and prototype, clustering, variance) and a between-subjects user study (N = 98) in which non-experts rated classification correctness, judged correctness as true/false, responded under time measurement, and reported helpfulness, trust, and satisfaction. The headline user-study result is a significant Explanation Type x Classification Type interaction on correctness ratings, F(1,98) = 9.11, p = .003, with FGNS participants rating misclassifications as less correct than k-NN participants did, while ratings for correct classifications were nearly identical. The paper also reports faster response times for FGNS and no significant differences in helpfulness, trust, or satisfaction.

Significance. If the user-study findings hold, FGNS would be a practical, low-cost method for supporting non-expert validation of model predictions in unfamiliar domains, and the paper addresses a real gap in example-based XAI. The manuscript has several strengths: it reports all five user-study measures including null results, it provides effect sizes, it includes a limitations section, and it clearly describes the algorithm pipeline. However, the quantitative evidence for RQ1 is largely circular because the analysis checks a property that FGNS optimizes by construction, and the judgment-accuracy measure rests on an undisclosed Likert-to-binary conversion. These issues weaken the support for the abstract's claim that FGNS produces 'more accurate decisions,' although the raw Likert-scale interaction is still a meaningful result. The central idea is defensible, but the current evaluation does not yet establish the stronger claim.

major comments (3)
  1. [§4.1, Eq. (2)] The Distance Analysis does not independently validate RQ1. Equation (2) defines FGNS selection as minimizing rho * sum_i ||M_i ⊙ (N_candidate - P_c)||^2, and Section 4.1 then shows that FGNS neighbors are closer to P_c under unweighted L2 distance (M = 4.14 vs. 5.55, t(298) = -19.37, p < .001). Because P_c appears directly in the selection objective, closer prototype distance is an expected consequence of the ranking rule rather than an empirical discovery that FGNS neighbors 'better reflect class characteristics.' The Cluster Analysis and Variance Analysis in the same section are restatements of the same distance-to-prototype comparison, not independent evidence. I recommend re-framing this analysis as a sanity check, or adding an external criterion such as expert-rated class representativeness or performance on held-out classes, to support RQ1.
  2. [§4.2, Judgment Accuracy] The binary True/False judgment measure is derived from the 5-point Likert correctness ratings, but the conversion rule is never specified. The paper calls the measure 'implicit' but does not state the threshold (e.g., whether ratings of 4 and 5 count as 'correct'), how neutral ratings of 3 are assigned, or whether the threshold was chosen before or after seeing the data. Because only aggregate means and standard deviations are reported, the reader cannot reconstruct the binary outcomes, and the reported interaction F(1,98) = 7.35, p = .008, as well as the means (True M = 0.70 vs. 0.65; False M = 0.17 vs. 0.21), may depend heavily on the chosen threshold. Since the abstract and discussion explicitly claim 'more accurate decisions,' the authors should report the exact binarization rule and provide a sensitivity analysis over plausible thresholds (e.g., >=4 vs. >3, with neutral ratings excluded or assigned to one category).
  3. [§4.2 and Table 1, Response Time] The response-time analysis reports F(1,93) whereas all other analyses use F(1,98), indicating that five participants or data points were excluded as 'outliers,' but the paper never states how many outliers were removed, by what criterion, or whether exclusion was decided before or after inspecting the data. The speed advantage of FGNS (M = 12.93s vs. 15.68s) is a secondary but non-negligible claim, and the current reporting does not allow the reader to assess whether the exclusion criterion materially affected the result. Please report the exclusion rule and the number of excluded observations.
minor comments (5)
  1. [§2.2] The reference formatting for 'L. Alfeo and Cimino [2024]' is inconsistent with the author-year citation style used throughout the rest of the paper.
  2. [§4.1, Figure 4] The y-axis of Figure 4 is labeled 'Frequency,' but the figure shows density-like distributions of Euclidean distances, and no units or binning details are provided; a box plot or density plot with labeled axes would be clearer.
  3. [§5] The discussion describes the lower False Judgment rate as reducing 'false positives,' but the operationalization counts incorrect correctness judgments, not false positives in the signal-detection sense; using 'false judgments' throughout would avoid confusion.
  4. [Algorithm 1 and §3.3] The symbol k is used both for the number of LIME superpixels and for the number of K-means clusters (k = 7); these are distinct quantities and should be renamed or explicitly differentiated.
  5. [General] The manuscript does not include a data or code availability statement; given that the judgment-accuracy measure depends on an undisclosed threshold, releasing anonymized item-level ratings would substantially aid reproducibility.

Circularity Check

1 steps flagged · score 6.0 of 10

Quantitative validation of FGNS tests its own optimization objective, making the RQ1 evidence circular; the user-study claim remains independent.

  1. self definitional [Section 3.3, Eq. (2); Section 4.1 Distance/Cluster/Variance Analyses]
    "FGNS ranks all training instances in class c using a feature-based distance metric: Lfeature = ρ Σ_i ||M_i ⊙ (N_candidate − P_c)||_2^2 ... Lower scores indicate stronger alignment with class-representative features. In contrast, when comparing distances to the predicted class prototype, FGNS neighbors were significantly closer (M = 4.14, IQR = 1.21) than those from k-NN (M = 5.55, IQR = 1.39), t(298) = −19.37, p < .001."

    FGNS is defined by exactly this loss: it ranks candidates by masked distance to the class prototype and selects the three lowest-loss neighbors. The Distance Analysis then 'finds' that FGNS-selected neighbors are closer to the prototype than k-NN's; this is a direct consequence of the selection rule, not an independent empirical discovery. The near-identical cluster result (M = 4.14 vs 5.55, SD = 0.73 vs 1.03) and lower variance (σ² = 0.53 vs 1.05) are corollaries of the same optimization objective: ranking by distance to the prototype concentrates selections near it. Thus the quantitative analyses offered for RQ1 confirm the method's construction rather than providing independent evidence of class-representativeness.

full rationale

The paper has two evidentiary pillars. RQ1's quantitative claims are circular: FGNS's selection rule (Eq. 2) minimizes masked distance to the class prototype, so showing selected neighbors are closer to the prototype, cluster more tightly around it, and have lower distance variance is checking the optimizer's own objective, not testing a separate hypothesis. This affects the Distance, Cluster, and Variance Analyses in Section 4.1 and the corresponding discussion. RQ2's user study is not circular: it compares human error-detection performance across explanation conditions using external behavioral data, so the headline claim that FGNS helps non-experts spot misclassifications has independent content. I do not find load-bearing self-citation: the COLE-HP and prior Ford/Keane citations motivate the twin-system setup but FGNS's ranking is newly defined, and no uniqueness theorem is imported. One non-circular but serious methodological limitation is flagged: the 'Judgment Accuracy' measure in Section 4.2 is called implicit and no Likert-to-binary threshold is specified, so the true/false judgment results are not reproducible as reported; this is an openness/reporting flaw, not a circularity. Overall, because one of the two central claims reduces to the method's own definition while the other stands independently, the appropriate score is 6 (partial circularity).

Assumptions & free parameters 6 free parameters · 5 assumptions · 1 invented entities

The central claim rests on design choices (k=7, rho=1.0, IoU cutoffs, sample sizes, undisclosed SAGE threshold) that are hand-set without ablations, plus domain assumptions about LIME, SAGE, and prototype well-formedness that the paper itself flags.

free parameters (6)
  • k (number of feature masks/clusters) = 7
    Set by hand in Section 3.3 and Algorithm 1 to balance feature diversity with efficiency; no sensitivity analysis given.
  • rho (mask weighting hyperparameter) = 1.0
    Set to equal mask weighting in Section 3.3; not derived or tuned against user performance.
  • LIME perturbation count = 500
    Algorithm 1 line 5; chosen without justification.
  • LIME aggregation sample size per class = 1,000
    Algorithm 1 line 3; arbitrary sample.
  • IoU thresholds = 0.5 (aggregation) and 0.8 (redundancy removal)
    Algorithm 1 lines 7 and 11; chosen by hand, no ablation.
  • SAGE global importance threshold = not specified (text says 'exceeding threshold')
    Algorithm 1 line 9; the threshold value is never stated, so the mask selection is not fully specified.
assumptions (5)
  • domain assumption LIME masks faithfully identify locally important superpixels for the CNN's prediction.
    Used in steps 1-2 of the FGNS pipeline (Section 3.1); the paper relies on LIME validity, which Huang and Marques-Silva 2023 critique.
  • domain assumption SAGE global importance scores correctly rank superpixel contributions to class-confidence.
    Used to filter masks in step 3; no calibration of SAGE to human perception is provided.
  • domain assumption The pixel-wise median of training images is a well-formed, human-meaningful class prototype.
    Section 3.2, Eq. 1; the authors themselves note in Section 5.1 that FGNS 'relies on interpretable superpixels and well-formed class prototypes', which may fail on natural images.
  • ad hoc to paper The 5-point Likert correctness rating can be converted to a binary True/False judgment with a single undisclosed threshold.
    The 'Judgment Accuracy' measure in Section 4.2 is called implicit, but the threshold is not reported; results depend on this choice.
  • standard math ANOVA assumptions (normality, sphericity) hold for the user-study measures.
    Two-way mixed ANOVAs are used in Section 4.2; the paper does not report assumption checks.
invented entities (1)
  • Class feature masks M_i (LIME+SAGE aggregated superpixel sets)
    purpose: Define class-representative image regions used to weight the neighbor-ranking loss in Eq. 2.
    The masks are constructed from LIME/SAGE on the training set; there is no independent behavioral test that these regions are what humans rely on when comparing digits.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Feature-Guided Neighbor Selection for Non-Expert Evaluation of Model Predictions." pith.science (2026). https://pith.science/paper/6BUFBMW5

@misc{pith2026250706029,
  author       = {Pith},
  title        = {Pith review of: Feature-Guided Neighbor Selection for Non-Expert Evaluation of Model Predictions},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/6BUFBMW5}},
  note         = {Machine review of arXiv:2507.06029}
}
read the original abstract

Explainable AI (XAI) methods often struggle to generate clear, interpretable outputs for users without domain expertise. We introduce Feature-Guided Neighbor Selection (FGNS), a post hoc method that enhances interpretability by selecting class-representative examples using both local and global feature importance. In a user study (N = 98) evaluating Kannada script classifications, FGNS significantly improved non-experts' ability to identify model errors while maintaining appropriate agreement with correct predictions. Participants made faster and more accurate decisions compared to those given traditional k-NN explanations. Quantitative analysis shows that FGNS selects neighbors that better reflect class characteristics rather than merely minimizing feature-space distance, leading to more consistent selection and tighter clustering around class prototypes. These results support FGNS as a step toward more human-aligned model assessment, although further work is needed to address the gap between explanation quality and perceived trust.

Figures

Figures reproduced from arXiv: 2507.06029 by the authors.

Figure 1
Figure 1. Visualization of LIME + SAGE feature integration. [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Illustration of the Kannada digit ”nalku” (4). (a) Ground- ¯ truth label; (b) class prototype formed via pixel-wise median; (c) atypical or noisy instance lacking a clear separation between the top curve and end stroke, which are features typically present in the class [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Example-based neighboring explanations for correct (left) [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Distribution of Euclidean distances for FGNS and k-NN neighbors. (a) Query-to-neighbor distances: FGNS neighbors are farther [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]
Figure 5
Figure 5. Figure 5: Proportion of True and False Judgments for FGNS and k [PITH_FULL_IMAGE:figures/full_fig_p006_5.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

28 extracted references · 25 canonical work pages

  1. [1]

    Proxy tasks and subjective measures can be misleading in eval- uating explainable ai systems

    [Buc ¸incaet al., 2020] Zana Buc ¸inca, Phoebe Lin, Krzysztof Z Gajos, and Elena L Glassman. Proxy tasks and subjective measures can be misleading in eval- uating explainable ai systems. In Proceedings of the 25th International Conference on Intelligent User Interfaces , pages 454–464,

    work page 2020

  2. [5]

    [Delaney et al., 2023] Eoin Delaney, Arjun Pakrashi, Derek Greene, and Mark T

    Curran Associates Inc. [Delaney et al., 2023] Eoin Delaney, Arjun Pakrashi, Derek Greene, and Mark T. Keane. Counterfactual explanations for misclassified images: How human and machine expla- nations differ. Artificial Intelligence, 324:103995,

    work page 2023

  3. [8]

    [Ewald et al., 2024] Fiona Katharina Ewald, Ludwig Both- mann, Marvin N

    Springer International Publishing. [Ewald et al., 2024] Fiona Katharina Ewald, Ludwig Both- mann, Marvin N. Wright, Bernd Bischl, Giuseppe Casal- icchio, and Gunnar K ¨onig. A guide to feature importance methods for scientific inference. In Luca Longo, Sebastian Lapuschkin, and Christin Seifert, editors, Explainable Ar- tificial Intelligence, pages 440–464, Cham,

    work page 2024

  4. [10]

    [Hoffman et al., 2018] Robert R Hoffman, Shane T Mueller, Gary Klein, and Jordan Litman

    Springer Nature Switzerland. [Hoffman et al., 2018] Robert R Hoffman, Shane T Mueller, Gary Klein, and Jordan Litman. Metrics for explain- able ai: Challenges and prospects. arXiv preprint arXiv:1812.04608,

    arXiv 2018

  5. [12]

    On two xai cultures: A case study of non-technical explana- tions in deployed ai system,

    [Jiang and Senge, 2021] Helen Jiang and Erwen Senge. On two xai cultures: A case study of non-technical explana- tions in deployed ai system,

    work page 2021

  6. [14]

    Explaining black-box classi- fiers using post-hoc explanations-by-example: The effect of explanations and error-rates in xai user studies

    [Kenny et al., 2021] Eoin M Kenny, Courtney Ford, Molly Quinn, and Mark T Keane. Explaining black-box classi- fiers using post-hoc explanations-by-example: The effect of explanations and error-rates in xai user studies. Artifi- cial Intelligence, 294:103459,

    work page 2021

  7. [15]

    Kenny, Mycal Tucker, and Julie Shah

    [Kenny et al., 2023] Eoin M. Kenny, Mycal Tucker, and Julie Shah. Towards interpretable deep reinforcement learning with human-friendly prototypes. In The Eleventh International Conference on Learning Representations ,

    work page 2023

  8. [16]

    Explainable artificial intelli- gence for human decision support system in the medical domain

    [Knapiˇc et al., 2021] Samanta Knapi ˇc, Avleen Malhi, Rohit Saluja, and Kary Fr ¨amling. Explainable artificial intelli- gence for human decision support system in the medical domain. Machine Learning and Knowledge Extraction , 3(3):740–770,

    work page 2021

Show all 28 references
  1. [17]

    Alfeo and Cimino, 2024] Antonio L

    [L. Alfeo and Cimino, 2024] Antonio L. Alfeo and Mario Cimino. Counterfactual-based feature importance for ex- plainable regression of manufacturing production quality measure. In Proceedings of the 13th International Confer- ence on Pattern Recognition Applications and Method...

    2024

  2. [18]

    From shap scores to feature importance scores,

    [Letoffe et al., 2024] Olivier Letoffe, Xuanxiang Huang, Nicholas Asher, and Joao Marques-Silva. From shap scores to feature importance scores,

    2024

  3. [19]

    Lundberg and Su-In Lee

    [Lundberg and Lee, 2017] Scott M. Lundberg and Su-In Lee. A unified approach to interpreting model predic- tions. In Proceedings of the 31st International Confer- ence on Neural Information Processing Systems, NIPS’17, page 4768–4777, Red Hook, NY , USA,

    2017

  4. [21]

    Expla- nation in human-ai systems: A literature meta-review, syn- opsis of key ideas and publications, and bibliography for explainable ai

    [Mueller et al., 2019] Shane T Mueller, Robert R Hoffman, William Clancey, Abigail Emrey, and Gary Klein. Expla- nation in human-ai systems: A literature meta-review, syn- opsis of key ideas and publications, and bibliography for explainable ai. arXiv preprint arXiv:1902.01876,

    2019 arXiv

  5. [22]

    ”why should i trust you?”: Ex- plaining the predictions of any classifier

    [Ribeiro et al., 2016] Marco Tulio Ribeiro, Sameer Singh, and Carlos Guestrin. ”why should i trust you?”: Ex- plaining the predictions of any classifier. In Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining , KDD ’16, page 1135–...

    2016

  6. [25]

    Searston and Ja- son M

    [Searston and Tangen, 2017] Rachel A. Searston and Ja- son M. Tangen. Expertise with unfamiliar objects is flexi- ble to changes in task but not changes in class.PLOS ONE, 12(6):1–14, 06

    2017

  7. [26]

    The human side of xai: Bridging the gap between ai and non-expert audiences

    [Severes et al., 2023] Beatriz Severes, Carolina Carreira, Ana Beatriz Vieira, Eduardo Gomes, Jo˜ao Tiago Apar´ıcio, and In ˆes Pereira. The human side of xai: Bridging the gap between ai and non-expert audiences. In Proceedings of the 41st ACM International Conference on Desi...

    2023

  8. [27]

    [Szymanski et al., 2021] Maxwell Szymanski, Martijn Mil- lecamp, and Katrien Verbert

    Association for Computing Machinery. [Szymanski et al., 2021] Maxwell Szymanski, Martijn Mil- lecamp, and Katrien Verbert. Visual, textual or hybrid: The effect of user expertise on different explanations. In 26th International Conference on Intelligent User Inter- faces, IUI ...

    2021

  9. [28]

    [Wang et al., 2019] Danding Wang, Qian Yang, Ashraf Ab- dul, and Brian Y

    Association for Computing Machinery. [Wang et al., 2019] Danding Wang, Qian Yang, Ashraf Ab- dul, and Brian Y . Lim. Designing theory-driven user- centric explainable ai. In Proceedings of the 2019 CHI Conference on Human Factors in Computing Systems , CHI ’19, page 1–15, New ...

    2019

  10. [1973]

    This looks like that: Deep learning for interpretable image recognition,

    [Chen et al., 2019] Chaofan Chen, Oscar Li, Chaofan Tao, Alina Jade Barnett, Jonathan Su, and Cynthia Rudin. This looks like that: Deep learning for interpretable image recognition,

    2019

  11. [2004]

    [Ehsan and Riedl, 2020] Upol Ehsan and Mark O. Riedl. Human-centered explainable ai: Towards a reflective sociotechnical approach. In Constantine Stephanidis, Masaaki Kurosu, Helmut Degen, and Lauren Reinerman- Jones, editors, HCI International 2020 - Late Breaking Papers: Mul...

    2020

  12. [2016]

    [Ridley, 2025] Michael Ridley

    Association for Computing Machinery. [Ridley, 2025] Michael Ridley. Human-centered explainable artificial intelligence: An annual review of information science and technology (arist) paper. Journal of the Asso- ciation for Information Science and Technology, 76(1):98– 120,

    2025

  13. [2017]

    [L¨ofstr¨om et al., 2024] Helena L ¨ofstr¨om, Tuwe L ¨ofstr¨om, Ulf Johansson, and Cecilia S ¨onstr¨od

    Curran As- sociates Inc. [L¨ofstr¨om et al., 2024] Helena L ¨ofstr¨om, Tuwe L ¨ofstr¨om, Ulf Johansson, and Cecilia S ¨onstr¨od. Calibrated explana- tions: With uncertainty information and counterfactuals. Expert Systems with Applications, 246:123154, July

    2024

  14. [2018]

    The inadequacy of shapley values for explainability,

    [Huang and Marques-Silva, 2023] Xuanxiang Huang and Joao Marques-Silva. The inadequacy of shapley values for explainability,

    2023

  15. [2019]

    Covert, Scott Lundberg, and Su- In Lee

    [Covert et al., 2020] Ian C. Covert, Scott Lundberg, and Su- In Lee. Understanding global feature contributions with additive importance measures. In Proceedings of the 34th International Conference on Neural Information Process- ing Systems, NIPS’20, Red Hook, NY , USA,

    2020

  16. [2020]

    Chase and Herbert A

    [Chase and Simon, 1973] William G. Chase and Herbert A. Simon. Perception in chess. Cognitive Psychology , 4(1):55–81,

    1973

  17. [2021]

    Twin-systems to explain artificial neural networks us- ing case-based reasoning: comparative tests of feature- weighting methods in ann-cbr twins for xai

    [Kenny and Keane, 2019] Eoin M Kenny and Mark T Keane. Twin-systems to explain artificial neural networks us- ing case-based reasoning: comparative tests of feature- weighting methods in ann-cbr twins for xai. In Proceed- ings of the 28th International Joint Conference on Arti...

    2019

  18. [2023]

    Explanation oriented retrieval

    [Doyle et al., 2004] D´onal Doyle, P ´adraig Cunningham, Derek Bridge, and Yusof Rahman. Explanation oriented retrieval. In European Conference on Case-Based Rea- soning, pages 157–168. Springer,

    2004

  19. [2024]

    [Ford and Keane, 2023] Courtney Ford and Mark T

    Springer Nature Switzerland. [Ford and Keane, 2023] Courtney Ford and Mark T. Keane. Explaining classifications to non-experts: An xai user study of post-hoc explanations for a classifier when people lack expertise. In Jean-Jacques Rousseau and Bill Kapra- los, editors, Patter...

    2023

  20. [2025]

    Stop explaining black box machine learning models for high stakes decisions and use interpretable models instead

    [Rudin, 2019] Cynthia Rudin. Stop explaining black box machine learning models for high stakes decisions and use interpretable models instead. Nature Machine Intelligence, 1(5):206–215,

    2019

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.