Pith. sign in

REVIEW 2 major objections 4 minor 73 references

From Pixels to Perception: Interpretable Predictions via Instance-wise Grouped Feature Selection

T0 review · 2 major / 4 minor · reviewed 2026-08-15 · deepseek-v4-flash

Pith's one-line read Region-based image masking yields interpretable predictions that rival black-box accuracy.

desk verdict A useful region-based feature-selection method whose faithfulness claim lacks a control for the mask-shape shortcut—worth reviewing, needs one fix. read the letter →

arxiv 2505.06003 v2 pith:GDZARFGX submitted 2025-05-09 cs.CV cs.LG

classification cs.CVcs.LG
keywords inherentinterpretabilityinstance-wisefeatureselectionsuperpixelsbinarymaskingdynamicthresholdingfaithfulnessinsertionfidelityimageclassification
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

P2P is a classifier that explains its own prediction by masking out most of the image, but it does so over superpixel regions rather than individual pixels. The paper argues that pixel-level sparsification is meaningless because evenly spaced or darkening masks preserve accuracy without removing information, whereas regions correspond to perceptual parts a human can grasp. P2P learns which regions to keep per instance, models relationships among parts, and dynamically lets the classifier request more regions until it reaches a user-set confidence level. The reported results aim to show that these region masks keep accuracy near a full black-box model while removing up to 80% of image content, locate target objects better than prior feature-selection methods, and are faithful in the sense that predictions actually depend on the kept regions.

What carries the argument

The machine is a region-level binary mask over SLIC superpixels. For each image, a frozen superpixel proposer partitions pixels into $D$ regions, an importance predictor outputs per-region logit-normal parameters $\mu$ and covariance $\Sigma$, with $\Sigma_{jk}=E_j\cdot E_k$ for learnable region embeddings so it is positive semi-definite; a Gumbel-Softmax sample binarizes the mask while remaining differentiable, and a thresholded sparsity loss $\mathcal{L}_m = -\log(1-\bar p)$ for $\bar p>\tau$ controls how many pixels remain. Dynamic thresholding at inference sets $\tau$ as the smallest value at which classifier confidence reaches $\delta$, making sparsity instance-specific.

What would settle it

Run P2P with a region proposer whose regions are random contiguous blobs of the same size distribution as the superpixels. If accuracy, localization, and insertion fidelity stay essentially unchanged, then the semantic meaningfulness of the regions is not doing the work; if they collapse, the superpixel premise is confirmed. Alternatively, a human annotation study can check whether the regions P2P keeps align with human object-part segmentations on the same images.

Watch

Extended reading notes

Core claim

The central claim is that inherently interpretable image predictions can be obtained by instance-wise grouped feature selection in the space of semantically meaningful regions. To carry this out, P2P freezes a superpixel proposer, predicts a selection probability per region from pixel-level features, draws a binary mask with the Gumbel-Softmax trick, and classifies only the unmasked regions; a non-diagonal logit-normal distribution over region probabilities encodes relationships among parts. Rather than fixing one sparsity level, P2P samples thresholds during training and at inference increases the threshold stepwise until the classifier's confidence reaches $\delta$, so the model effectively asks for more evidence when needed. On CIFAR-10, COCO-10, ImageNet, and ImageNet-9 the paper reports accuracy comparable to the upper-bounding black-box while retaining only 20–50% of pixels depending on dataset, the best localization among baselines on COCO-10 and ImageNet-9, and the steepest insertion and deletion fidelity curves, which it reads as evidence that predictions are based on the selected regions and hence inherently interpretable.

Load-bearing premise

That SLIC superpixels partition an image into perceptually meaningful atomic regions, so that switching a region off removes one understandable feature; if superpixels instead split or merge object parts arbitrarily, P2P's masks would be no more meaningful than pixel masks.

Editorial extensions

If this is right

  • Pixel-level instance-wise feature selection is shown to be insufficient: evenly spaced masks that remove 80% of pixels keep near-black-box accuracy, so optimization must be carried out over regions to enforce informative sparsity.
  • If P2P's results hold, an interpretable classifier can be both faithful and accurate, which weakens the usual accuracy-interpretability trade-off for this setup.
  • Dynamic thresholding gives users a direct control: lowering the confidence threshold $\delta$ produces sparser explanations, and raising it yields more evidence, so sparsity can be tuned per application without retraining.
  • Steep insertion and deletion curves mean faithfulness can be demonstrated empirically for masked-input models, providing a template for evaluating future inherently interpretable vision methods.
  • Visualizing the kept regions exposes potential dataset shortcuts such as pillows or teddy bears signaling 'bed', making the method a tool for detecting spurious correlations in datasets.

Reading between the lines

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

  • The paper never directly tests whether the superpixel proposer matches human part decomposition; a user study asking humans to name or segment the retained regions would settle whether the interpretability claim is perceptual or only structural.
  • Because the paper's ablation only swaps SLIC for Watershed, it does not separate grouping from semantic meaning; comparing P2P against random contiguous region partitions of matched granularity would isolate whether the regions' content, not just their contiguity, drives the gains.
  • The same grouped-feature-selection recipe should transfer to audio or text, where atomic segments such as phonemes, words, or spans play the role of superpixels, offering a direct test of the method's generality.
  • The embedding-based covariance could be read as a part-object scene graph; clustering the embeddings instead of coloring them would let P2P output named part groupings, an extension the paper only gestures at.
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

2 major / 4 minor

Summary. The paper proposes P2P, an instance-wise feature-selection method for image classification that makes predictions from a sparse binary mask over superpixel regions. A frozen SLIC proposer partitions each image; a selector predicts per-region selection parameters, modeled jointly through a logit-normal distribution with a covariance matrix built from learnable embeddings; Gumbel-Softmax sampling yields a differentiable binary mask; and a classifier predicts from the masked input. The method uses a thresholded sparsity loss and a dynamic inference-time threshold that increases the retained fraction until classifier certainty exceeds a user-set value. Experiments on CIFAR-10, ImageNet, ImageNet-9, COCO-10, and the BAM datasets compare accuracy, localization, and insertion/deletion fidelity against DiET, REAL-X, RB-AEM, B-cos, COMET, and control baselines. The paper reports accuracy comparable to a black-box classifier, the best localization on COCO-10 and ImageNet-9, and the steepest fidelity curves, and concludes that P2P is inherently interpretable.

Significance. If the central claims are borne out, P2P is a useful contribution: it demonstrates that grouping pixels into regions and modeling their joint selection can produce sparse, object-focused masks with accuracy competitive with full-image classifiers, and the dynamic thresholding idea directly addresses the fixed-sparsity limitation of prior feature-selection methods. The evaluation is thorough in several respects: results are reported over ten seeds with standard deviations, code is released, the COMET-1 control gives a concrete faithfulness check for COMET, the BAM semi-synthetic datasets provide a localization check with known ground truth, and ablations cover the superpixel algorithm and the certainty threshold. The proof of positive semi-definiteness of the covariance in Appendix A is elementary and correct. However, the faithfulness evidence does not exclude a mask-pattern shortcut, and the perceptual meaningfulness of the superpixel grouping is asserted rather than validated; both issues bear directly on the headline interpretability claim.

major comments (2)
  1. [Section 5, Figures 3 and 7; Section 2] The claim that P2P faithfully uses the selected regions is not established because there is no control for the mask-pattern shortcut. Since the classifier is trained jointly with the selector and always receives x_m = m⊙x, it can predict from the spatial arrangement of zero and nonzero regions; a ViT can exploit the silhouette of the mask as a class cue. Insertion and deletion fidelity (Figures 3 and 7) cannot separate content-based from pattern-based prediction because both mechanisms make the curves steep. The paper itself identifies this failure mode for REAL-X in Section 2 and notes that REAL-X counters it with random masking, but P2P adopts no analogous safeguard. Please add one or more of the following controls and report results: (i) train or evaluate a classifier on the binary mask alone; (ii) replace the contents of the selected regions with class-independent noise or a constant and measure accuracy; (iii) train P2P with random masking and compare. If either control shows that the mask pattern alone drives accuracy, the central faithfulness claim in Sections 5 and 6 should be withdrawn or substantially weakened.
  2. [Section 3; Appendix B, Table 5] The premise that SLIC superpixels are 'perceptually meaningful atomic regions' is not validated. Section 3 relies on this premise to argue that region-level masking yields human-understandable explanations, but the only related experiment, Appendix B Table 5, swaps SLIC for Watershed and finds similar accuracy and localization; this tests algorithm invariance, not whether the regions correspond to the parts humans use for recognition. Please provide a direct check, for example overlap with human part or object annotations or a user study comparing the interpretability of P2P masks against pixel-level masks, or explicitly narrow the claim to 'contiguous regions' rather than 'perceptually meaningful parts.'
minor comments (4)
  1. [Appendix B, Table 3] On BAM Scene, COMET achieves slightly higher localization (98.49 ± 0.63) than P2P (98.18 ± 0.27); the sentence 'Clearly, P2P also excels in this controlled setup' should acknowledge this exception or report the numbers explicitly.
  2. [Section 5; Table 4] The statement that P2P removes 'up to 80%' of image content is an average over certain datasets; for ImageNet the average masked fraction is about 52% (Table 4). Please make the dataset-specific nature of this claim explicit.
  3. [Section 3; Section 5] The claimed benefit of the logit-normal covariance for selection is not isolated. Please add an ablation with a diagonal covariance, or no covariance, to show how much of the accuracy or localization gain comes from modeling part relationships.
  4. [Section 4; Table 1] The sparsity levels τ reported in Table 1 appear to be rounded averages of P2P's dynamic masks (compare Table 4), while the text describes them as fixed budgets. Please clarify how τ is set for P2P and for each baseline so the comparison is unambiguous.

Circularity Check

1 steps flagged · score 6.0 of 10

Faithfulness evidence in P2P is partly self-definitional: the insertion-fidelity curve is forced by thresholding the selection probabilities that define the mask.

  1. self definitional [Section 4 (Metrics; Implementation Details) and Section 5 (Faithfulness)]
    "Insertion starts with a dark image, iteratively adding the most important pixels of xm. In both metrics, we compute the fidelity of the predictions with respect to the original prediction on xm, i.e. using \hat y as target rather than y. ... At inference, we set the certainty threshold δ to 0.8 for ImageNet and 0.99 for all other datasets, and determine active regions by thresholding probabilities at 0.5 instead of sampling."

    P2P's active mask is defined as the set of regions whose selection probability exceeds 0.5. Insertion fidelity orders pixels by the same selection probabilities in descending order. Therefore all active regions are inserted before any inactive region, and at the sparsity fraction τ the reconstructed image is exactly x_m. Since the fidelity target is the model's prediction on x_m, the insertion fidelity at τ equals the model's confidence on its own masked input, which is at least δ by the dynamic-thresholding construction. Hence the steep insertion curve, cited as evidence that 'the explanations output by P2P are the actual reasons for the prediction,' is a formal artifact of thresholding rather than an independent demonstration that the selected regions cause the prediction.

full rationale

P2P's accuracy and localization results are benchmarked against external targets (test accuracy, ground-truth segmentations), so those pillars are not circular. The dynamic-thresholding procedure is also carefully designed to avoid peeking at the predicted class. However, the faithfulness pillar is partly self-definitional: because the mask is the thresholded selection probability and insertion fidelity ranks by that same probability against the prediction on x_m, the insertion curve must reach the model's confidence at the sparsity level by construction. This affects a central claim of the paper, but it is only one of three evaluation pillars and does not reduce the accuracy or localization results to the method's own definitions. A minor self-citation for the logit-normal parameterization appears but is not load-bearing.

Assumptions & free parameters 6 free parameters · 5 assumptions · 0 invented entities

P2P introduces no new physical entities. The part-specific embeddings and the dynamic thresholding rule are architectural components of the method, not independent postulates with external falsifiable handles.

free parameters (6)
  • lambda_1 (sparsity loss weight) = 10
    Hand-set to a large value so that any sufficiently large weight drives the expected sparsity toward the threshold tau; not tuned per dataset.
  • lambda_2 (covariance regularizer) = 0.01
    Hand-set to avoid overfitting in the covariance term; not tuned per dataset.
  • Number of superpixels per image = 100
    Chosen by visual inspection in Section 4; ablation in Appendix B shows no strong dependence on the superpixel algorithm, but the segment count is a free design choice.
  • SLIC compactness parameter m = 20
    FastSLIC hyperparameter, reported in Section 4 as part of the region proposer configuration.
  • Certainty threshold delta = 0.8 (ImageNet), 0.99 (others)
    User-set confidence level that controls the dynamic sparsity; ablation over delta is provided in Appendix B, Table 4.
  • Selection probability threshold at inference = 0.5
    Active regions are determined by thresholding the learned selection probabilities at 0.5 rather than sampling; this is a hand-chosen binarization rule.
assumptions (5)
  • domain assumption SLIC superpixels correspond to perceptually meaningful atomic regions
    Section 1 argues humans perceive objects as parts; Section 3 uses FastSLIC as the frozen region proposer. No human study or perceptual metric validates this correspondence.
  • domain assumption The classifier's softmax confidence is a reliable measure of prediction certainty for threshold selection
    Equation (4) uses max_c p(y_c | x_m) >= delta to decide when enough regions have been selected. The paper provides an ablation over delta but does not test whether confidence tracks correctness or information sufficiency.
  • domain assumption Insertion/deletion fidelity computed on models trained with masking is a valid faithfulness measure
    Section 4 and Appendix B argue that training with masked pixels keeps the masked inputs in-distribution, making ROAR-style retraining unnecessary; this is an argument, not a proof.
  • standard math The covariance matrix defined by dot products of embeddings is positive semi-definite
    Appendix A proves PSD via z^T Sigma z = ||sum z_j E_j||^2 >= 0.
  • standard math Gumbel-Softmax provides a differentiable approximation to sampling binary masks
    Standard result (Jang et al., 2017; Maddison et al., 2017) used in Section 3 for backpropagation through the mask sampling.

how reviews work

0 comments
Cite this review

Pith. "Pith review of From Pixels to Perception: Interpretable Predictions via Instance-wise Grouped Feature Selection." pith.science (2026). https://pith.science/paper/GDZARFGX

@misc{pith2026250506003,
  author       = {Pith},
  title        = {Pith review of: From Pixels to Perception: Interpretable Predictions via Instance-wise Grouped Feature Selection},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/GDZARFGX}},
  note         = {Machine review of arXiv:2505.06003}
}
read the original abstract

Understanding the decision-making process of machine learning models provides valuable insights into the task, the data, and the reasons behind a model's failures. In this work, we propose a method that performs inherently interpretable predictions through the instance-wise sparsification of input images. To align the sparsification with human perception, we learn the masking in the space of semantically meaningful pixel regions rather than on pixel-level. Additionally, we introduce an explicit way to dynamically determine the required level of sparsity for each instance. We show empirically on semi-synthetic and natural image datasets that our inherently interpretable classifier produces more meaningful, human-understandable predictions than state-of-the-art benchmarks.

Figures

Figures reproduced from arXiv: 2505.06003 by the authors.

Figure 1
Figure 1. Masking 80% of input under different constraints. All 3 masks lead to similar predictive performance but only P2P provides interpretability by sparsity. the context of instance-wise feature selection. Moreover, the proposed approach is equipped to model the relationship across parts to capture part-object relations. Lastly, we address the limitation of a fixed level of sparsity and propose a dynamic thresholding tha… view at source ↗
Figure 2
Figure 2. Schematic overview of P2P. A frozen region proposer partitions the input x into perceptually meaningful parts R1:D, which are assigned a learned selection probability. The mask is then binarized by sampling, leading to xm that serves as the input to the classifier. As described in the previous paragraph, the region-wise parameters are predicted at pixel-level and then aggregated within each region. To ensure that th… view at source ↗
Figure 3
Figure 3. Insertion Fidelity, where the most important pixels of the explanation xm are iteratively added to a black image, measuring how much information is required until the original prediction is recovered. The faster, i.e. the steeper the curve, the better. Results are reported as averages and standard deviations across ten seeds. Ablation Study: Fixing τ A key contribution of P2P is its dynamic thresholding based on the… view at source ↗
Figures from the paper (11 more)
Figure 4
Figure 4. Figure 4: Ablation study of P2P with a constant τ . In “Matched”, we set τ to the average value of the dynamic P2P, which is 20%, 20%, 40%, 30%, for each dataset, respectively. Faithfulness From these previous results, we can deduce that P2P provides both meaningful explanations…
Figure 5
Figure 5. Figure 5: Masked inputs xm of COCO-10 for selected methods. We also see that the P2P variant with a fixed τ is worse than the dynamic P2P for low insertion percentages p but better when p > τ . This is expected, as dynamic thresholding results in varying levels of masking across…
Figure 6
Figure 6. Figure 6: Insertion Fidelity, where the most important pixels of the explanation xm are iteratively added to a black image, measuring how much information is required until the original prediction is recovered. The faster, i.e. the steeper the curve, the better. Results are repo…
Figure 7
Figure 7. Figure 7: Deletion Fidelity, where the most important pixels of the explanation xm are iteratively removed, measuring how much information needs to be removed until the prediction changes. The steeper the curve, the better. Results are reported as averages and standard deviation…
Figure 8
Figure 8. Figure 8: Visualization of the part-wise embeddings, used to compute the covariance, as colors. Left is the partitioned input, middle is the masked input, and on the right, we show the part embeddings. Randomly Sampled Masks We believe that one of the best ways to assess and und…
Figure 9
Figure 9. Figure 9: Visualization of randomly sampled images overlayed with their partitions, as well as the masked input xm for BAM Object. 19 [PITH_FULL_IMAGE:figures/full_fig_p019_9.png]
Figure 10
Figure 10. Figure 10: Visualization of randomly sampled images overlayed with their partitions, as well as the masked input xm for BAM Scene. 20 [PITH_FULL_IMAGE:figures/full_fig_p020_10.png]
Figure 11
Figure 11. Figure 11: Visualization of randomly sampled images overlayed with their partitions, as well as the masked input xm for CIFAR-10. 21 [PITH_FULL_IMAGE:figures/full_fig_p021_11.png]
Figure 12
Figure 12. Figure 12: Visualization of randomly sampled images overlayed with their partitions, as well as the masked input xm for COCO-10. 22 [PITH_FULL_IMAGE:figures/full_fig_p022_12.png]
Figure 13
Figure 13. Figure 13: Visualization of randomly sampled images overlayed with their partitions, as well as the masked input xm for ImageNet-9. 23 [PITH_FULL_IMAGE:figures/full_fig_p023_13.png]
Figure 14
Figure 14. Figure 14: Visualization of randomly sampled images overlayed with their partitions, as well as the masked input xm for ImageNet. 24 [PITH_FULL_IMAGE:figures/full_fig_p024_14.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

73 extracted references · 38 canonical work pages

  1. [1]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION format.date year duplicate empty "emp...

  2. [2]

    Slic superpixels compared to state-of-the-art superpixel methods

    Achanta, R., Shaji, A., Smith, K., Lucchi, A., Fua, P., and S \"u sstrunk, S. Slic superpixels compared to state-of-the-art superpixel methods. IEEE transactions on pattern analysis and machine intelligence, 34 0 (11): 0 2274--2282, 2012

  3. [3]

    Sanity checks for saliency maps

    Adebayo, J., Gilmer, J., Muelly, M., Goodfellow, I., Hardt, M., and Kim, B. Sanity checks for saliency maps. Advances in neural information processing systems, 31, 2018

  4. [4]

    B-cosification: Transforming deep neural networks to be inherently interpretable

    Arya, S., Rao, S., Boehle, M., and Schiele, B. B-cosification: Transforming deep neural networks to be inherently interpretable. In 38th Conference on Neural Information Processing Systems, 2024

  5. [5]

    and Shen, S

    Atchison, J. and Shen, S. M. Logistic-normal distributions: Some properties and uses. Biometrika, 67 0 (2): 0 261--272, 1980

  6. [6]

    Meditationes sacrae

    Bacon, F. Meditationes sacrae. Excusum impensis Humfredi Hooper., 1597. URL https://books.google.ch/books?id=HUyeAQAACAAJ

  7. [7]

    Discriminative feature attributions: bridging post hoc explainability and inherent interpretability

    Bhalla, U., Srinivas, S., and Lakkaraju, H. Discriminative feature attributions: bridging post hoc explainability and inherent interpretability. Advances in Neural Information Processing Systems, 36, 2024

  8. [8]

    Recognition-by-components: a theory of human image understanding

    Biederman, I. Recognition-by-components: a theory of human image understanding. Psychological review, 94 0 (2): 0 115, 1987

Show all 73 references
  1. [9]

    B-cos networks: Alignment is all we need for interpretability

    B \"o hle, M., Fritz, M., and Schiele, B. B-cos networks: Alignment is all we need for interpretability. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp.\ 10329--10338, 2022

  2. [10]

    B-cos alignment for inherently interpretable cnns and vision transformers

    B \"o hle, M., Singh, N., Fritz, M., and Schiele, B. B-cos alignment for inherently interpretable cnns and vision transformers. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2024

  3. [11]

    Carballo-Castro, A., Laguna, S., Vandenhirtz, M., and Vogt, J. E. Exploiting interpretable capabilities with concept-enhanced diffusion and prototype networks. arXiv preprint arXiv:2410.18705, 2024

  4. [12]

    Chen, C., Li, O., Tao, D., Barnett, A., Rudin, C., and Su, J. K. This looks like that: deep learning for interpretable image recognition. Advances in neural information processing systems, 32, 2019

  5. [13]

    Learning to explain: An information-theoretic perspective on model interpretation

    Chen, J., Song, L., Wainwright, M., and Jordan, M. Learning to explain: An information-theoretic perspective on model interpretation. In International conference on machine learning, pp.\ 883--892. PMLR, 2018

  6. [14]

    J., Lee, S., Chun, S., Akata, Z., and Shim, H

    Choe, J., Oh, S. J., Lee, S., Chun, S., Akata, Z., and Shim, H. Evaluating weakly supervised object localization methods right. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pp.\ 3133--3142, 2020

  7. [15]

    C., Qiu, W., Lu, M., Kim, N

    Covert, I. C., Qiu, W., Lu, M., Kim, N. Y., White, N. J., and Lee, S.-I. Learning to maximize mutual information for dynamic feature selection. In International Conference on Machine Learning, pp.\ 6424--6447. PMLR, 2023

  8. [16]

    and Kim, B

    Doshi-Velez, F. and Kim, B. Towards a rigorous science of interpretable machine learning. arXiv preprint arXiv:1702.08608, 2017

  9. [17]

    and Kim, B

    Doshi-Velez , F. and Kim, B. Towards A Rigorous Science of Interpretable Machine Learning , March 2017

  10. [18]

    Concept embedding models: Beyond the accuracy-explainability trade-off

    Espinosa Zarlenga, M., Barbiero, P., Ciravegna, G., Marra, G., Giannini, F., Diligenti, M., Shams, Z., Precioso, F., Melacci, S., Weller, A., et al. Concept embedding models: Beyond the accuracy-explainability trade-off. In Advances in Neural Information Processing Systems, vo...

  11. [19]

    Scaling rectified flow transformers for high-resolution image synthesis

    Esser, P., Kulal, S., Blattmann, A., Entezari, R., M \"u ller, J., Saini, H., Levi, Y., Lorenz, D., Sauer, A., Boesel, F., et al. Scaling rectified flow transformers for high-resolution image synthesis. In Forty-first International Conference on Machine Learning, 2024

  12. [20]

    Fanconi, C., Vandenhirtz, M., Husmann, S., and Vogt, J. E. This reads like that: Deep learning for interpretable natural language processing. In The 2023 Conference on Empirical Methods in Natural Language Processing, 2023

  13. [21]

    Shapley values for feature selection: The good, the bad, and the axioms

    Fryer, D., Str \"u mke, I., and Nguyen, H. Shapley values for feature selection: The good, the bad, and the axioms. Ieee Access, 9: 0 144352--144360, 2021

  14. [22]

    Interpretations steered network pruning via amortized inferred saliency maps

    Ganjdanesh, A., Gao, S., and Huang, H. Interpretations steered network pruning via amortized inferred saliency maps. In European Conference on Computer Vision, pp.\ 278--296. Springer, 2022

  15. [23]

    Geirhos, R., Jacobsen, J.-H., Michaelis, C., Zemel, R., Brendel, W., Bethge, M., and Wichmann, F. A. Shortcut learning in deep neural networks. Nature Machine Intelligence, 2 0 (11): 0 665--673, 2020

  16. [24]

    A benchmark for interpretability methods in deep neural networks

    Hooker, S., Erhan, D., Kindermans, P.-J., and Kim, B. A benchmark for interpretability methods in deep neural networks. Advances in neural information processing systems, 32, 2019

  17. [25]

    Searching for mobilenetv3

    Howard, A., Sandler, M., Chu, G., Chen, L.-C., Chen, B., Tan, M., Wang, W., Zhu, Y., Pang, R., Vasudevan, V., et al. Searching for mobilenetv3. In Proceedings of the IEEE/CVF international conference on computer vision, pp.\ 1314--1324, 2019

  18. [26]

    On the concept trustworthiness in concept bottleneck models

    Huang, Q., Song, J., Hu, J., Zhang, H., Wang, Y., and Song, M. On the concept trustworthiness in concept bottleneck models. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 38, pp.\ 21161--21168, 2024

  19. [27]

    and Goldberg, Y

    Jacovi, A. and Goldberg, Y. Towards faithfully interpretable nlp systems: How should we define and evaluate faithfulness? arXiv preprint arXiv:2004.03685, 2020

  20. [28]

    Categorical reparameterization with gumbel-softmax

    Jang, E., Gu, S., and Poole, B. Categorical reparameterization with gumbel-softmax. In 5th International Conference on Learning Representations, ICLR 2017, Toulon, France, April 24-26, 2017, Conference Track Proceedings . OpenReview.net, 2017. URL https://openreview.net/forum?...

  21. [29]

    Have we learned to explain?: How interpretability methods can learn to encode predictions in their interpretations

    Jethani, N., Sudarshan, M., Aphinyanaphongs, Y., and Ranganath, R. Have we learned to explain?: How interpretability methods can learn to encode predictions in their interpretations. In International Conference on Artificial Intelligence and Statistics, pp.\ 1459--1467. PMLR, 2021

  22. [30]

    Fast slic: Efficient superpixel segmentation, 2021

    Kim, A. Fast slic: Efficient superpixel segmentation, 2021. URL https://github.com/Algy/fast-slic. Accessed: 2025-01-25

  23. [31]

    Kim, B., Rudin, C., and Shah, J. A. The bayesian case model: A generative approach for case-based reasoning and prototype classification. Advances in neural information processing systems, 27, 2014

  24. [32]

    Kingma, D. P. and Ba, J. Adam: A method for stochastic optimization. In Bengio, Y. and LeCun, Y. (eds.), 3rd International Conference on Learning Representations, ICLR 2015, San Diego, CA, USA, May 7-9, 2015, Conference Track Proceedings , 2015. URL http://arxiv.org/abs/1412.6980

  25. [33]

    W., Nguyen, T., Tang, Y

    Koh, P. W., Nguyen, T., Tang, Y. S., Mussmann, S., Pierson, E., Kim, B., and Liang, P. Concept bottleneck models. In III, H. D. and Singh, A. (eds.), Proceedings of the 37th International Conference on Machine Learning, volume 119 of Proceedings of Machine Learning Research, p...

  26. [34]

    Learning multiple layers of features from tiny images

    Krizhevsky, A., Hinton, G., et al. Learning multiple layers of features from tiny images. 2009

  27. [35]

    N., Sun, J., Cetin, N., Al-Hazwani, I., Schlegel, U., Cheng, F., and El-Assady, M

    Laguna, S., Heidenreich, J. N., Sun, J., Cetin, N., Al-Hazwani, I., Schlegel, U., Cheng, F., and El-Assady, M. Explimeable: A visual analytics approach for exploring lime. In 2023 Workshop on Visual Analytics in Healthcare (VAHC), pp.\ 27--33. IEEE, 2023

  28. [36]

    R., Dale, D., Elsahar, H., Heffernan, K., a o Maria Janeiro, J., Tran, T., Ropers, C., Sánchez, E., Roman, R

    LCM team , Barrault, L., Duquenne, P.-A., Elbayad, M., Kozhevnikov, A., Alastruey, B., Andrews, P., Coria, M., Couairon, G., Costa-juss\` a , M. R., Dale, D., Elsahar, H., Heffernan, K., a o Maria Janeiro, J., Tran, T., Ropers, C., Sánchez, E., Roman, R. S., Mourachko, A., Sal...

  29. [37]

    Lin, T.-Y., Maire, M., Belongie, S., Hays, J., Perona, P., Ramanan, D., Doll \'a r, P., and Zitnick, C. L. Microsoft coco: Common objects in context. In Computer Vision--ECCV 2014: 13th European Conference, Zurich, Switzerland, September 6-12, 2014, Proceedings, Part V 13, pp....

  30. [38]

    Lipton, Z. C. The Mythos of Model Interpretability . Communications of the ACM, 61 0 (10): 0 35--43, June 2016. doi:10.48550/arxiv.1606.03490

  31. [39]

    Lipton, Z. C. The doctor just won't accept that! arXiv preprint arXiv:1711.08037, 2017

  32. [40]

    Rotating features for object discovery

    L \"o we, S., Lippe, P., Locatello, F., and Welling, M. Rotating features for object discovery. Advances in Neural Information Processing Systems, 36, 2024

  33. [41]

    A unified approach to interpreting model predictions

    Lundberg, S. A unified approach to interpreting model predictions. arXiv preprint arXiv:1705.07874, 2017

  34. [42]

    Interpretable image classification with adaptive prototype-based vision transformers

    Ma, C., Donnelly, J., Liu, W., Vosoughi, S., Rudin, C., and Chen, C. Interpretable image classification with adaptive prototype-based vision transformers. In The Thirty-eighth Annual Conference on Neural Information Processing Systems, 2024

  35. [43]

    J., Mnih, A., and Teh, Y

    Maddison, C. J., Mnih, A., and Teh, Y. W. The concrete distribution: A continuous relaxation of discrete random variables. In 5th International Conference on Learning Representations, ICLR 2017, Toulon, France, April 24-26, 2017, Conference Track Proceedings . OpenReview.net, ...

  36. [44]

    and Vogt, J

    Marcinkevi c s, R. and Vogt, J. E. Interpretable and explainable machine learning: a methods-centric overview with concrete examples. Wiley Interdisciplinary Reviews: Data Mining and Knowledge Discovery, 13 0 (3): 0 e1493, 2023

  37. [45]

    Marcinkevi c s, R., Laguna, S., Vandenhirtz, M., and Vogt, J. E. Beyond concept bottleneck models: How to make black boxes intervenable? In Advances in neural information processing systems, volume 37, 2024

  38. [46]

    Miller, G. A. The magical number seven, plus or minus two: Some limits on our capacity for processing information. Psychological review, 63 0 (2): 0 81, 1956

  39. [47]

    Stochastic segmentation networks: Modelling spatially correlated aleatoric uncertainty

    Monteiro, M., Le Folgoc, L., Coelho de Castro, D., Pawlowski, N., Marques, B., Kamnitsas, K., van der Wilk, M., and Glocker, B. Stochastic segmentation networks: Modelling spatially correlated aleatoric uncertainty. In Advances in neural information processing systems, volume ...

  40. [48]

    and Protzel, P

    Neubert, P. and Protzel, P. Compact watershed and preemptive slic: On improving trade-offs of superpixel segmentation algorithms. In 2014 22nd international conference on pattern recognition, pp.\ 996--1001. IEEE, 2014

  41. [49]

    Local feature selection without label or feature leakage for interpretable machine learning predictions

    Oosterhuis, H., Lyu, L., and Anand, A. Local feature selection without label or feature leakage for interpretable machine learning predictions. arXiv preprint arXiv:2407.11778, 2024

  42. [50]

    Palmer, S. E. Hierarchical structure in perceptual representation. Cognitive psychology, 9 0 (4): 0 441--474, 1977

  43. [51]

    Rise: Randomized input sampling for explanation of black-box models

    Petsiuk, V. Rise: Randomized input sampling for explanation of black-box models. arXiv preprint arXiv:1806.07421, 2018

  44. [52]

    and Bolon-Canedo, V

    Remeseiro, B. and Bolon-Canedo, V. A review of feature selection methods in medical applications. Computers in biology and medicine, 112: 0 103375, 2019

  45. [53]

    why should i trust you?

    Ribeiro, M. T., Singh, S., and Guestrin, C. " why should i trust you?" explaining the predictions of any classifier. In Proceedings of the 22nd ACM SIGKDD international conference on knowledge discovery and data mining, pp.\ 1135--1144, 2016

  46. [54]

    Metaphysics

    Ross, W. Metaphysics. CreateSpace Independent Publishing Platform, 2016. ISBN 9781534926288. URL https://books.google.ch/books?id=aVQbvwEACAAJ

  47. [55]

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

    Rudin, C. Stop explaining black box machine learning models for high stakes decisions and use interpretable models instead. Nature machine intelligence, 1 0 (5): 0 206--215, 2019

  48. [56]

    Imagenet large scale visual recognition challenge

    Russakovsky, O., Deng, J., Su, H., Krause, J., Satheesh, S., Ma, S., Huang, Z., Karpathy, A., Khosla, A., Bernstein, M., et al. Imagenet large scale visual recognition challenge. International journal of computer vision, 115: 0 211--252, 2015

  49. [57]

    A review of feature selection techniques in bioinformatics

    Saeys, Y., Inza, I., and Larranaga, P. A review of feature selection techniques in bioinformatics. bioinformatics, 23 0 (19): 0 2507--2517, 2007

  50. [58]

    R., Cogswell, M., Das, A., Vedantam, R., Parikh, D., and Batra, D

    Selvaraju, R. R., Cogswell, M., Das, A., Vedantam, R., Parikh, D., and Batra, D. Grad-cam: Visual explanations from deep networks via gradient-based localization. In Proceedings of the IEEE international conference on computer vision, pp.\ 618--626, 2017

  51. [59]

    Learning important features through propagating activation differences

    Shrikumar, A., Greenside, P., and Kundaje, A. Learning important features through propagating activation differences. In International conference on machine learning, pp.\ 3145--3153. PMlR, 2017

  52. [60]

    T., Dosovitskiy, A., Brox, T., and Riedmiller, M

    Springenberg, J. T., Dosovitskiy, A., Brox, T., and Riedmiller, M. Striving for simplicity: The all convolutional net. arXiv preprint arXiv:1412.6806, 2014

  53. [61]

    Axiomatic attribution for deep networks

    Sundararajan, M., Taly, A., and Yan, Q. Axiomatic attribution for deep networks. In International conference on machine learning, pp.\ 3319--3328. PMLR, 2017

  54. [62]

    S., Mrabti, F., and Zahi, A

    Tadist, K., Najah, S., Nikolov, N. S., Mrabti, F., and Zahi, A. Feature selection methods and genomic big data: a systematic review. Journal of Big Data, 6 0 (1): 0 1--24, 2019

  55. [63]

    A., Havaei, M., Berthier, T., Dutil, F., Di Jorio, L., Hamarneh, G., and Bengio, Y

    Taghanaki, S. A., Havaei, M., Berthier, T., Dutil, F., Di Jorio, L., Hamarneh, G., and Bengio, Y. Infomask: Masked variational latent representation to localize chest disease. In Medical Image Computing and Computer Assisted Intervention--MICCAI 2019: 22nd International Confer...

  56. [64]

    Regression shrinkage and selection via the lasso

    Tibshirani, R. Regression shrinkage and selection via the lasso. Journal of the Royal Statistical Society Series B: Statistical Methodology, 58 0 (1): 0 267--288, 1996

  57. [65]

    Training data-efficient image transformers & distillation through attention

    Touvron, H., Cord, M., Douze, M., Massa, F., Sablayrolles, A., and J \'e gou, H. Training data-efficient image transformers & distillation through attention. In International conference on machine learning, pp.\ 10347--10357. PMLR, 2021

  58. [66]

    Vandenhirtz, M., Laguna, S., Marcinkevi c s, R., and Vogt, J. E. Stochastic concept bottleneck models. In The Thirty-eighth Annual Conference on Neural Information Processing Systems, 2024

  59. [67]

    Y., Engstrom, L., Ilyas, A., and Madry, A

    Xiao, K. Y., Engstrom, L., Ilyas, A., and Madry, A. Noise or signal: The role of image backgrounds in object recognition. In International Conference on Learning Representations, 2021

  60. [68]

    and Kim, B

    Yang, M. and Kim, B. Benchmarking attribution methods with relative feature importance. arXiv preprint arXiv:1907.09701, 2019

  61. [69]

    Invase: Instance-wise variable selection using neural networks

    Yoon, J., Jordon, J., and Van der Schaar, M. Invase: Instance-wise variable selection using neural networks. In International conference on learning representations, 2018

  62. [70]

    and Lin, Y

    Yuan, M. and Lin, Y. Model selection and estimation in regression with grouped variables. Journal of the Royal Statistical Society Series B: Statistical Methodology, 68 0 (1): 0 49--67, 2006

  63. [71]

    Comprehensive attribution: Inherently explainable vision model with feature detector

    Zhang, X., Lee, D., and Wang, S. Comprehensive attribution: Inherently explainable vision model with feature detector. In European Conference on Computer Vision, pp.\ 196--213. Springer, 2025

  64. [72]

    A survey on evaluation methods for image segmentation

    Zhang, Y.-J. A survey on evaluation methods for image segmentation. Pattern recognition, 29 0 (8): 0 1335--1346, 1996

  65. [73]

    Places: A 10 million image database for scene recognition

    Zhou, B., Lapedriza, A., Khosla, A., Oliva, A., and Torralba, A. Places: A 10 million image database for scene recognition. IEEE transactions on pattern analysis and machine intelligence, 40 0 (6): 0 1452--1464, 2017

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.