Naming the Concepts Classifiers Rely On: Language-Anchored Decomposition for Faithful Explanation

Ahsan Habib Akash; Binod Bhattarai; Dipkamal Bhusal; Donald A. Adjeroh; Prashnna Kumar Gyawali; Stacey Jones

REVIEW 2 major objections 6 minor 29 references

Naming the Concepts Classifiers Rely On: Language-Anchored Decomposition for Faithful Explanation

T0 review · 2 major / 6 minor · reviewed 2026-07-09 · glm-5.2

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

Pith's one-line read Named concepts, faithful explanations, no retraining

desk verdict Writing letter on LAD paper read the letter →

arxiv 2607.07264 v1 pith:ZV4MJSME submitted 2026-07-08 cs.CV

Ahsan Habib Akash , Dipkamal Bhusal , Stacey Jones , Donald A. Adjeroh , Binod Bhattarai , Prashnna Kumar Gyawali This is my paper · ORCID

classification cs.CV

keywords concept-basedexplanationnon-negativematrixfactorizationCLIPvision-languagemodelspost-hocinterpretabilityfaithfulnessmedicalimaging

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 introduces Language-Anchored Decomposition (LAD), a method for explaining what a trained image classifier relies on when making a decision. Existing explanation methods face a trade-off: those that recover factors faithful to the model's internal behavior leave those factors unnamed, while those that attach human-readable concept names require retraining or altering the classifier. LAD breaks this trade-off by inverting a standard matrix factorization technique. Instead of learning both the concept coefficients and the concept basis from the data, LAD fixes the coefficients to spatial similarity maps derived from a vision-language model (CLIP), where each map corresponds to a named concept proposed by a large language model. It then learns only the concept basis that best reconstructs the frozen classifier's own activations. This makes naming a structural constraint rather than a post-hoc label, while the model's own feature geometry determines which named concepts survive. The paper shows that this language anchor is functional: removing it preserves classification accuracy but collapses the faithfulness of attributions, demonstrating that language supervision determines which concept directions are discovered. Across natural images, scenes, and medical imaging, LAD produces spatially precise, named concepts that are decision-relevant under insertion and deletion tests.

What carries the argument

The key machinery is the inversion of non-negative matrix factorization: fixing a language-grounded coefficient matrix S (built from CLIP spatial similarity maps for LLM-proposed named concepts) and learning only a non-negative concept basis W that reconstructs the frozen encoder's spatial activations via projected gradient descent. At inference, non-negative least squares estimates per-image concept coefficients against the fixed basis.

What would settle it

If replacing the fixed language-grounded coefficient matrix with learned coefficients (standard unsupervised NMF) produced equivalent or better deletion faithfulness (C-Del), the central claim that the language anchor is functional rather than cosmetic would be falsified. The paper reports the opposite: C-Del drops from 0.902 to 0.507 when the anchor is removed.

Watch

Extended reading notes

Core claim

The central mechanism is the inversion of non-negative matrix factorization (NMF). Standard NMF decomposes a model's activations into a coefficient matrix and a basis matrix, both learned from data, producing unnamed factors. LAD fixes the coefficient matrix S to CLIP-derived spatial similarity maps for named concepts and learns only the basis W against the frozen encoder's activations. This means each basis vector is structurally tied to a named concept from the outset, and the reconstruction objective reshapes initial CLIP similarities so that concepts inconsistent with the encoder's evidence are suppressed. The paper's key empirical finding is that this language anchor changes which basis

Load-bearing premise

The method assumes that CLIP-based localized similarity maps, constructed via red-circle visual prompting on a grid, provide a semantically meaningful and spatially accurate coefficient matrix that can anchor the decomposition. If these CLIP maps are noisy or do not align with the encoder's actual feature geometry, the learned basis may not correspond to faithful concepts.

Editorial extensions

If this is right

If the language anchor genuinely steers decomposition toward classifier-meaningful structure, then concept-level auditing of deployed models becomes possible without retraining, enabling domain experts to inspect not just where a model looks but what named evidence it attributes to those regions.
The finding that removing the language anchor preserves accuracy but collapses deletion faithfulness suggests that accuracy alone is an insufficient proxy for explanation quality; faithfulness metrics may require semantic grounding to be meaningful.
The clinical domain results, where named concepts anchor to focal disease evidence and lead deletion faithfulness by a wide margin, suggest the method may be most valuable in domains where discriminative evidence is spatially localized and clinically named.
The negative result on Vision Transformers, where reconstruction transfers but spatial localization fails, indicates that the approach's success depends on the encoder's spatial geometry and that transformer-native formulations are needed for patch-token architectures.
The boundary bias observed in dermoscopy, where concepts localize to lesion borders rather than diagnostic interiors, reveals that named concept faithfulness is domain-dependent and that the method may surface where the model's evidence sits rather than asserting clinically complete reasoning.

Reading between the lines

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

The dependence on CLIP-derived similarity maps as the fixed coefficient matrix means LAD's concept quality is bounded by CLIP's spatial grounding fidelity; if CLIP's red-circle prompting mislocalizes a concept, the learned basis will be anchored to the wrong region, a risk the paper acknowledges but does not systematically quantify across concept types.
The observation that the language anchor changes which basis directions are discovered suggests a broader principle: any externally fixed coefficient structure, not necessarily language-derived, could steer factorization toward specific semantic regimes, opening the question of whether non-linguistic priors (e.g., anatomical atlases, spectral signatures) could serve as alternative anchors for doma
The trade-off between naming completeness and deletion faithfulness (LAD leads insertion but sometimes trails on deletion) implies that semantically constrained explanations may systematically under-cover decision-relevant but non-interpretable evidence, raising the question of whether the unnamed residual factors represent genuinely non-conceptual computation or simply concepts outside the LLM's

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

The main thing to know: this paper inverts NMF in a way I haven't seen before—fixing CLIP-derived language similarity maps as the coefficient matrix S and learning only the basis W against frozen encoder activations. That gives you named concepts without retraining the model, which is a real gap in the interpretability literature that nobody had cleanly filled. The per-image named-concept rollout (Figure 4) is a genuine new capability; baselines can only emit “factor 1, factor 2.” Code is public, which matters here because the method is simple enough to reproduce and check yourself. The clinical results on retinal fundus are strong—LAD leads C-Del by a wide margin (0.954 vs. 0.863 for FACE)—and the computational cost is an order of magnitude below the baselines. The paper is also unusually honest about where the method fails, which I respect. The ViT case study (Sec. A7) is a clean negative result: reconstruction transfers but localization collapses, and they explain why. The dermoscopy boundary bias (Sec. A5) is reported as a limitation, not hidden. Now the soft spots. The stress-test note lands on a real gap: the “named” pillar of the central claim rests on qualitative evidence (Figures 3–5) and one cross-image consistency example. C-Ins and C-Del test decision-relevance, not whether the name attached to a concept actually describes what's in the highlighted region. No human study, no automated naming-accuracy metric. That's a hole. The ablation in Table 3 is the other weak point. It compares fixed CLIP-derived S against learned S (standard NMF), but this conflates two changes: fixing S vs. learning S, and using CLIP-derived semantic content vs. unconstrained values. A random fixed S—matched in sparsity and scale—might also improve C-Del over learned S, because any structured coefficient matrix constrains the optimization. Without that control, the ablation shows that fixing S helps, not that the language content specifically helps. The paper claims the anchor is “functional, not cosmetic,” but the evidence supports “fixing S helps,” which is weaker. These are evaluation gaps, not method flaws. The core formulation is sound, the convergence analysis is clean, and the honest reporting of failures builds trust. The paper deserves a serious referee who can push for a random-anchor control and some quantitative naming validation. I'd recommend accepting for peer review.

Referee Report

2 major / 6 minor

Summary. The paper proposes Language-Anchored Decomposition (LAD), a post-hoc concept-discovery method that inverts standard NMF by fixing CLIP-derived language-grounded similarity maps as the coefficient matrix S and learning only the concept basis W against frozen encoder activations. The central claim is that LAD produces concepts that are simultaneously named, faithful, and obtained without modifying the model. The method is evaluated on ImageNet (ResNet34, 500 classes), Places365 (ResNet50, 364 classes), and two clinical domains (retinal fundus on ODIR-5K with DenseNet-121, dermoscopy on HAM10000 with ConvNeXt-Small). The paper provides an ablation (Table 3) showing that removing the language anchor preserves accuracy but collapses deletion faithfulness (C-Del drops from 0.902 to 0.507 on ImageNet). Code is publicly available.

Significance. The paper addresses a genuine gap in the interpretability literature: post-hoc concept methods (ACE, ICE, CRAFT, FACE) recover faithful but unnamed factors, while by-design methods (Concept Bottleneck Models, LaViSE) attach names only by retraining or altering the classifier. LAD's approach of inverting NMF to fix language-grounded coefficients and learn only the basis is a clean and novel formulation. The convergence analysis (Sec. A3) is correct: with S fixed, the objective is a convex quadratic in W, and projected gradient descent converges to a global minimizer. The computational efficiency of fixing S (0.03s/class vs. 0.5s for CRAFT and 2.6s for FACE) is a practical advantage. The ViT case study (Sec. A7) is a commendable negative result that honestly documents where the method does not transfer. The publicly available code and the breadth of evaluation (natural images, scenes, two medical domains, five CNN backbones, one ViT) strengthen reproducibility.

major comments (2)

[Sec. 5.5, Table 3] The ablation comparing LAD (fixed CLIP-derived S) against unsupervised NMF (learned S) is the paper's primary evidence that the language anchor is 'functional, not cosmetic.' However, this comparison conflates two changes: (1) fixing S vs. learning S, and (2) using CLIP-derived semantic values vs. unconstrained values. A random or structured-random fixed S (matched in sparsity and scale to the CLIP maps) might also improve C-Del over fully learned S, because fixing S regularizes the decomposition regardless of semantic content. To isolate whether the language content of S matters — as opposed to the structural constraint of fixing S — the paper should include a control where S is fixed but semantically arbitrary (e.g., random non-negative matrices or shuffled CLIP maps). If such a control also achieves high C-Del, the claim that 'language supervision determines which concept directions'
[Central claim (Abstract, Sec. 1, Sec. 6)] The central claim includes three pillars: named, faithful, and no modification. The 'faithful' pillar is tested quantitatively via C-Ins and C-Del (Table 1). The 'no modification' pillar is structural. But the 'named' pillar — that concept names accurately describe the visual content of the localized regions — is supported only qualitatively (Figures 3-5) and by one cross-image consistency example (Figure 4, 'maple neck'). C-Ins and C-Del measure whether the decomposition captures decision-relevant directions, not whether the names attached to those directions are correct. A concept could achieve high C-Del while having a misleading name if CLIP mislocalizes the concept. The paper should acknowledge this gap explicitly and, ideally, provide a quantitative or semi-quantitative evaluation of naming accuracy (e.g., human ratings on a sample of concept-region pairs, or automated verification

minor comments (6)

[Table 1] LAD's C-Del on ImageNet (0.902) is lower than FACE (0.939). The paper frames this as an intended trade-off (Sec. 5.2: 'Trading a portion of deletion completeness for named, verifiable concepts'). This framing is reasonable but could be stated more prominently, as readers may initially read the table as LAD underperforming.
[Sec. 3.3] The red-circle visual prompting technique for constructing CLIP similarity maps is a key component, but the choice of circle radius r and grid resolution (7x7 or 14x14) is not thoroughly justified or analyzed. Table A6 shows grid sensitivity for cost but not for faithfulness metrics. A brief analysis of how grid resolution affects C-Ins/C-Del would strengthen the method.
[Sec. A5, Table A2] The dermoscopy results show substantially lower C-Del (0.418) compared to retinal fundus (0.954). The paper attributes this to boundary bias in concept localization (Fig. A1). This is an honest and important finding, but it raises a question about the generality of the method in domains where discriminative features are interior rather than boundary-localized. A brief discussion of when LAD is expected to succeed or struggle would help practitioners.
[Sec. A7] The ViT case study reports C-Del = 0.02, which the paper attributes to the preserved [CLS] pathway and diffuse patch-token localization. This is a valuable negative result. The paper could note more explicitly that the near-zero C-Del is partly an artifact of the evaluation protocol (the [CLS] token carries original information), making the metric not directly comparable to the CNN setting.
[Fig. 2] The diagram is informative but dense. The relationship between the CLIP-derived S matrix and the learned W basis could be made clearer with a small worked example showing dimensions for a single image.
[Sec. 3.2] The claim that 'naming becomes a structural property of the factorization rather than a post-hoc step' is a strong statement. It would be more precise to say naming is enforced by the structure of S, since the names are still ultimately derived from CLIP similarity and could be incorrect if CLIP's localization is poor.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity; derivation chain is self-contained with one minor self-citation that is not load-bearing.

full rationale

The paper's method chain is: (1) LLM generates concept vocabulary (external input), (2) CLIP produces spatial similarity maps S (external, via red-circle prompting), (3) basis W is learned to minimize ||Ā − SW^T||_F against frozen encoder activations (Eq. 1), (4) at inference, Ŝ is estimated via non-negative least squares (Eq. 3), (5) faithfulness is measured by C-Ins and C-Del (standard metrics applied uniformly to all baselines). No step reduces to its own inputs by construction. The reconstruction accuracy (Acc) is somewhat expected to be high since W is optimized for reconstruction, but the paper explicitly acknowledges this ('All methods preserve accuracy, so faithfulness is read from concept insertion and deletion') and does not claim Acc as a novel result—it is a prerequisite checked across all methods. The C-Ins and C-Del metrics test decision-relevance independently of the fitting objective. The ablation (Table 3) comparing fixed CLIP-derived S vs. learned S provides independent evidence that the anchor changes the discovered basis directions. The skeptic's concern that the ablation conflates 'fixing S' with 'using CLIP-derived semantic values' is a valid experimental design critique, but it is a correctness/evaluation concern, not circularity. The self-citation to FACE [3] (co-authored by Bhusal) is not load-bearing: FACE is used as a baseline comparator, not as a premise or theorem justifying LAD's central claim. No uniqueness theorem, no ansatz smuggled via self-citation, no fitted parameter renamed as prediction. The 'named' property resting on qualitative evidence is an evaluation completeness gap, not a circularity in the derivation. Score 1 reflects the minor, non-load-bearing self-citation and the acknowledged (but non-central) reconstruction-accuracy circularity shared with all baselines.

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

The method relies on existing tools (CLIP, LLMs, NMF) and does not invent new entities. The free parameters are standard hyperparameters with sensitivity analyses.

free parameters (3)

r (concept budget) = 25
Number of concepts per class, chosen to match baselines. Sensitivity analysis in Table 4.
Probe grid size = 7x7 or 14x14
Spatial resolution of CLIP probing, chosen based on dataset (Sec. A6).
CLIP similarity threshold = 0.80
Used in vocabulary filtering to remove near-duplicates (Sec. A1.4).

assumptions (3)

domain assumption CLIP image-text similarity maps provide spatially localized semantic information that can anchor concept decomposition.
This is the core assumption underlying the construction of S. It is invoked in Sec. 3.3 and validated empirically on CNNs but shown to fail on ViTs (Sec. A7).
domain assumption Red-circle visual prompting effectively localizes concepts in CLIP.
Used in Sec. 3.3 to construct spatial maps. Relies on prior work [21].
domain assumption LLMs can generate diverse, visually grounded concept vocabularies.
Used in Sec. 3.3 to generate the concept set T_c.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Naming the Concepts Classifiers Rely On: Language-Anchored Decomposition for Faithful Explanation." pith.science (2026). https://pith.science/paper/ZV4MJSME

@misc{pith2026260707264,
  author       = {Pith},
  title        = {Pith review of: Naming the Concepts Classifiers Rely On: Language-Anchored Decomposition for Faithful Explanation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ZV4MJSME}},
  note         = {Machine review of arXiv:2607.07264}
}

read the original abstract

Deep neural networks are widely deployed in high-stakes visual applications where interpretability is critical, yet existing explanations face a trade-off: post-hoc concept methods recover factors that are faithful to a model's behavior but unnamed, while naming and by-design methods attach human-readable concepts only by retraining or altering the classifier. We propose Language-Anchored Decomposition (LAD), a post-hoc framework that delivers concepts which are simultaneously named, faithful, and obtained without modifying the model. For each class, a large language model proposes a concept vocabulary that CLIP-based similarity maps localize across image regions. Inverting standard non-negative matrix factorization, LAD fixes these language-grounded maps as the coefficient matrix and learns only a concept basis that reconstructs the frozen encoder's activations, so naming becomes a structural constraint and the model's own feature geometry determines which concepts are retained. Removing this anchor preserves accuracy but collapses attribution faithfulness. Across natural-image, scene, and medical-imaging benchmarks, LAD produces spatially precise explanations that are decision-relevant under both concept insertion and deletion, while uniquely providing stable, human-interpretable concept names.

Figures

Figures reproduced from arXiv: 2607.07264 by the authors.

**Figure 2.** LAD framework. (a) An input image is processed through a pretrained encoder f(·) to obtain latent activation maps, divided into spatial patches and arranged into a matrix A¯ ∈ R N×p . (b) Using CLIP-based semantic priors, the concept similarity matrix S is fixed as the coefficient matrix while the basis W is learned through a language-anchored reconstruction A¯ ≈ SW⊤; the reconstructed activations preserve spatial c… view at source ↗

**Figure 3.** Concept discovery results on ImageNet. The upper panel shows top-ranked concepts from NMF baselines (ICE, CRAFT, FACE), [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗

Figures from the paper (3 more)

**Figure 4.** Figure 4: Per-image named-concept rollouts, a capability unique to LAD. Top: for an electric guitar (p=0.99) and an axolotl (p=1.00), LAD returns localized, named concepts with their reconstruction contribution cˆ (Eq. (7)); FACE returns factors of comparable contribution but wi…

**Figure 5.** Figure 5: Named-concept rollouts on retinal fundus images. For a diabetic retinopathy case (top) and an age-related macular degeneration case (bottom), LAD decomposes the prediction into separately named, localized concepts. Each map is anchored to a clinical term, so the explan…

**Figure 6.** Figure 6: Concept activation scores before and after optimization, across 50 images. Before optimization, scores largely reflect the CLIP [PITH_FULL_IMAGE:figures/full_fig_p009_6.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

29 extracted references · 29 canonical work pages

[1]

Network dissection: Quantifying inter- pretability of deep visual representations

David Bau, Bolei Zhou, Aditya Khosla, Aude Oliva, and Antonio Torralba. Network dissection: Quantifying inter- pretability of deep visual representations. InProceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2017. 2

work page 2017
[2]

Show and tell: Visually explainable deep neural nets via spatially-aware concept bot- tleneck models

Itay Benou and Tammy Riklin Raviv. Show and tell: Visually explainable deep neural nets via spatially-aware concept bot- tleneck models. InProceedings of the Computer Vision and Pattern Recognition Conference, pages 30063–30072, 2025. 2, 3

work page 2025
[3]

Face: Faithful automatic concept extraction

Dipkamal Bhusal, Michael Clifford, Sara Rampazzi, and Nidhi Rastogi. Face: Faithful automatic concept extraction. NeurIPS, 2025. 1, 2, 5

work page 2025
[4]

Svd based ini- tialization: A head start for nonnegative matrix factorization

Christos Boutsidis and Efstratios Gallopoulos. Svd based ini- tialization: A head start for nonnegative matrix factorization. Pattern Recognition, 41(4):1350–1362, 2008. 4, 12

work page 2008
[5]

What i cannot predict, i do not understand: A human- centered evaluation framework for explainability methods

Julien Colin, Thomas Fel, R ´emi Cad`ene, and Thomas Serre. What i cannot predict, i do not understand: A human- centered evaluation framework for explainability methods. Advances in neural information processing systems, 35: 2832–2845, 2022. 2

work page 2022
[6]

Craft: Concept recursive activation factoriza- tion for explainability

Thomas Fel, Agustin Picard, Louis Bethune, Thibaut Boissin, David Vigouroux, Julien Colin, R ´emi Cad`ene, and Thomas Serre. Craft: Concept recursive activation factoriza- tion for explainability. InProceedings of the IEEE/CVF con- ference on computer vision and pattern recognition, pages 2711–2721, 2023. 1, 2, 5

work page 2023
[7]

Towards automatic concept-based explanations.Ad- vances in neural information processing systems, 32, 2019

Amirata Ghorbani, James Wexler, James Y Zou, and Been Kim. Towards automatic concept-based explanations.Ad- vances in neural information processing systems, 32, 2019. 1, 2

work page 2019
[8]

Natu- ral language descriptions of deep visual features

Evan Hernandez, Sarah Schwettmann, David Bau, Teona Bagashvili, Antonio Torralba, and Jacob Andreas. Natu- ral language descriptions of deep visual features. InInter- national Conference on Learning Representations (ICLR),

work page

Show all 29 references

[9]

Interpretability be- yond feature attribution: Quantitative testing with concept activation vectors (tcav)

Been Kim, Martin Wattenberg, Justin Gilmer, Carrie Cai, James Wexler, Fernanda Viegas, et al. Interpretability be- yond feature attribution: Quantitative testing with concept activation vectors (tcav). InInternational conference on ma- chine learning, pages 2668–2677. PMLR, 2018. 1, 2

2018
[10]

Concept bottleneck models

Pang Wei Koh, Thao Nguyen, Yew Siang Tang, Stephen Mussmann, Emma Pierson, Been Kim, and Percy Liang. Concept bottleneck models. InInternational conference on machine learning, pages 5338–5348. PMLR, 2020. 2

2020
[11]

The mythos of model interpretability: In machine learning, the concept of interpretability is both important and slippery.Queue, 16(3):31–57, 2018

Zachary C Lipton. The mythos of model interpretability: In machine learning, the concept of interpretability is both important and slippery.Queue, 16(3):31–57, 2018. 5

2018
[12]

Interpretability beyond classification output: Semantic bottleneck networks

Max Losch, Mario Fritz, and Bernt Schiele. Interpretability beyond classification output: Semantic bottleneck networks. arXiv preprint arXiv:1907.10882, 2019. 2

1907 arXiv
[13]

General pitfalls of model-agnostic interpretation methods for machine learning models

Christoph Molnar, Gunnar K ¨onig, Julia Herbinger, Timo Freiesleben, Susanne Dandl, Christian A Scholbeck, Giuseppe Casalicchio, Moritz Grosse-Wentrup, and Bernd Bischl. General pitfalls of model-agnostic interpretation methods for machine learning models. InInternational Work...

2020
[14]

The ef- fectiveness of feature attribution methods and its correlation with automatic evaluation scores.Advances in Neural Infor- mation Processing Systems, 34:26422–26436, 2021

Giang Nguyen, Daeyoung Kim, and Anh Nguyen. The ef- fectiveness of feature attribution methods and its correlation with automatic evaluation scores.Advances in Neural Infor- mation Processing Systems, 34:26422–26436, 2021. 1, 2 8 Figure 6. Concept activation scores before and ...

2021
[15]

CLIP-Dissect: Au- tomatic description of neuron representations in deep vision networks

Tuomas Oikarinen and Tsui-Wei Weng. CLIP-Dissect: Au- tomatic description of neuron representations in deep vision networks. InInternational Conference on Learning Repre- sentations (ICLR), 2023. 2

2023
[16]

Label-free concept bottleneck models.Inter- national Conference on Learning Representations (ICLR),

Tuomas Oikarinen, Subhro Das, Lam M Nguyen, and Tsui- Wei Weng. Label-free concept bottleneck models.Inter- national Conference on Learning Representations (ICLR),
[17]

Rise: Random- ized input sampling for explanation of black-box models

Vitali Petsiuk, Abir Das, and Kate Saenko. Rise: Random- ized input sampling for explanation of black-box models. InProceedings of the British Machine Vision Conference (BMVC), 2018. 2

2018
[18]

Learning transferable visual models from natural language supervi- sion

Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, et al. Learning transferable visual models from natural language supervi- sion. InInternational conference on machine learning, p...

2021
[19]

Stop explaining black box machine learn- ing models for high stakes decisions and use interpretable models instead.Nature machine intelligence, 1(5):206–215,

Cynthia Rudin. Stop explaining black box machine learn- ing models for high stakes decisions and use interpretable models instead.Nature machine intelligence, 1(5):206–215,
[20]

Grad-cam: Visual explanations from deep networks via gradient-based localization

Ramprasaath R Selvaraju, Michael Cogswell, Abhishek Das, Ramakrishna Vedantam, Devi Parikh, and Dhruv Batra. Grad-cam: Visual explanations from deep networks via gradient-based localization. InProceedings of the IEEE in- ternational conference on computer vision, pages 618–626,
[21]

What does clip know about a red circle? vi- sual prompt engineering for vlms

Aleksandar Shtedritski, Christian Rupprecht, and Andrea Vedaldi. What does clip know about a red circle? vi- sual prompt engineering for vlms. InProceedings of the IEEE/CVF International Conference on Computer Vision, pages 11987–11997, 2023. 3

2023
[22]

Deep inside convolutional networks: Visualising image clas- sification models and saliency maps

Karen Simonyan, Andrea Vedaldi, and Andrew Zisserman. Deep inside convolutional networks: Visualising image clas- sification models and saliency maps. In2nd International Conference on Learning Representations (ICLR), 2014. 2

2014
[23]

Smoothgrad: removing noise by adding noise.arXiv preprint arXiv:1706.03825, 2017

Daniel Smilkov, Nikhil Thorat, Been Kim, Fernanda Vi ´egas, and Martin Wattenberg. Smoothgrad: removing noise by adding noise.arXiv preprint arXiv:1706.03825, 2017. 2

2017 arXiv
[24]

Axiomatic attribution for deep networks

Mukund Sundararajan, Ankur Taly, and Qiqi Yan. Axiomatic attribution for deep networks. InInternational conference on machine learning, pages 3319–3328. PMLR, 2017. 1, 2

2017
[25]

Discovering and mitigating biases in clip-based image editing

Md Mehrab Tanjim, Krishna Kumar Singh, Kushal Kafle, Ritwik Sinha, and Garrison W Cottrell. Discovering and mitigating biases in clip-based image editing. InProceed- ings of the IEEE/CVF Winter Conference on Applications of Computer Vision, pages 2984–2993, 2024. 6

2024
[26]

Learning bottleneck concepts in image classification

Bowen Wang, Liangzhi Li, Yuta Nakashima, and Hajime Na- gahara. Learning bottleneck concepts in image classification. InProceedings of the ieee/cvf conference on computer vision and pattern recognition, pages 10962–10971, 2023. 2

2023
[27]

Explain- ing deep convolutional neural networks via latent visual- semantic filter attention

Yu Yang, Seungbae Kim, and Jungseock Joo. Explain- ing deep convolutional neural networks via latent visual- semantic filter attention. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2022. 2

2022
[28]

{class_name}

Ruihan Zhang, Prashan Madumal, Tim Miller, Krista A. Ehinger, and Benjamin I. P. Rubinstein. Invertible concept- based explanations for cnn models with non-negative con- cept activation vectors. InProceedings of the AAAI Confer- ence on Artificial Intelligence, 2021. 1, 2, 5 9...

2021
[29]

skin irregularity

We then define a uniform grid ofHt ×W t spatial loca- tions (typically7×7). For each grid cell center, we overlay a thin red circular outline with fixed radiusron the image. This produces a set ofH tWt modified images: {Ip}HtWt p=1 , where eachI p highlights a different spatia...

Pith tools

Reviewed July 9, 2026 · model on record in the stance chip above.

[1] [1]

Network dissection: Quantifying inter- pretability of deep visual representations

David Bau, Bolei Zhou, Aditya Khosla, Aude Oliva, and Antonio Torralba. Network dissection: Quantifying inter- pretability of deep visual representations. InProceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2017. 2

work page 2017

[2] [2]

Show and tell: Visually explainable deep neural nets via spatially-aware concept bot- tleneck models

Itay Benou and Tammy Riklin Raviv. Show and tell: Visually explainable deep neural nets via spatially-aware concept bot- tleneck models. InProceedings of the Computer Vision and Pattern Recognition Conference, pages 30063–30072, 2025. 2, 3

work page 2025

[3] [3]

Face: Faithful automatic concept extraction

Dipkamal Bhusal, Michael Clifford, Sara Rampazzi, and Nidhi Rastogi. Face: Faithful automatic concept extraction. NeurIPS, 2025. 1, 2, 5

work page 2025

[4] [4]

Svd based ini- tialization: A head start for nonnegative matrix factorization

Christos Boutsidis and Efstratios Gallopoulos. Svd based ini- tialization: A head start for nonnegative matrix factorization. Pattern Recognition, 41(4):1350–1362, 2008. 4, 12

work page 2008

[5] [5]

What i cannot predict, i do not understand: A human- centered evaluation framework for explainability methods

Julien Colin, Thomas Fel, R ´emi Cad`ene, and Thomas Serre. What i cannot predict, i do not understand: A human- centered evaluation framework for explainability methods. Advances in neural information processing systems, 35: 2832–2845, 2022. 2

work page 2022

[6] [6]

Craft: Concept recursive activation factoriza- tion for explainability

Thomas Fel, Agustin Picard, Louis Bethune, Thibaut Boissin, David Vigouroux, Julien Colin, R ´emi Cad`ene, and Thomas Serre. Craft: Concept recursive activation factoriza- tion for explainability. InProceedings of the IEEE/CVF con- ference on computer vision and pattern recognition, pages 2711–2721, 2023. 1, 2, 5

work page 2023

[7] [7]

Towards automatic concept-based explanations.Ad- vances in neural information processing systems, 32, 2019

Amirata Ghorbani, James Wexler, James Y Zou, and Been Kim. Towards automatic concept-based explanations.Ad- vances in neural information processing systems, 32, 2019. 1, 2

work page 2019

[8] [8]

Natu- ral language descriptions of deep visual features

Evan Hernandez, Sarah Schwettmann, David Bau, Teona Bagashvili, Antonio Torralba, and Jacob Andreas. Natu- ral language descriptions of deep visual features. InInter- national Conference on Learning Representations (ICLR),

work page