Pith. sign in

REVIEW 3 major objections 4 minor 46 references

Towards Better Generalization and Interpretability in Unsupervised Concept-Based Models

T0 review · 3 major / 4 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read Unsupervised concept-based image classifiers can stay interpretable and still nearly match black-box accuracy.

desk verdict LCBM is a genuinely new unsupervised concept-based model with strong empirical work, but the paper's headline accuracy claims are overstated and need scoping. read the letter →

arxiv 2506.02092 v1 pith:Z74BDFDW submitted 2025-06-02 cs.LG cs.AIstat.ML

Francesco De Santis , Philippe Bich , Gabriele Ciravegna , Pietro Barbiero , Danilo Giordano , Tania Cerquitelli This is my paper
classification cs.LGcs.AIstat.ML
keywords concept-basedmodelsunsupervisedlearninginterpretabilityXAIvariationalinferenceconceptembeddingsimageclassificationBernoullilatentspace
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 LCBM, an unsupervised concept-based model for image classification that learns concepts as Bernoulli random variables in a latent space, each paired with a learned vector embedding. Its central claim is that this embedding-based concept representation closes most of the accuracy gap that has historically separated unsupervised interpretable models from black-box classifiers: across seven image datasets LCBM's task accuracy is within 1-2% of an end-to-end black-box baseline and often higher than prior unsupervised concept models. The paper also claims the learned concepts retain more information about both input and label, align better with human-annotated concepts, and are judged more plausible and understandable by users. If these claims hold, an interpretable model no longer has to sacrifice accuracy to explain itself, and concept annotations from humans or large language models become unnecessary.

What carries the argument

The load-bearing mechanism is the concept embedding pair $(c_j, t_j)$: for each concept $j$, a per-concept MLP $h_j$ maps a frozen backbone's features into a $d=128$ vector $c_j$, and a learned prototype $t_j$ converts alignment $c_j \cdot t_j$ through a sigmoid into a Bernoulli activation probability, sampled via the reparameterization trick to give the concept score. The same embedding feeds both the decoder for reconstruction and a class-specific network that predicts local linear weights, so each prediction stays a linear combination of concept scores while the representations carry more than one bit per concept. The ELBO objective with batch-level KL regularization ties representativity, completeness, and alignment together.

What would settle it

Construct a capacity-matched baseline by giving SENN or BotCL the same 128-dimensional embedding per concept (and a comparable parameter count) while keeping their original losses and interpretable linear head; if its task accuracy reaches LCBM's 1-2% gap on CIFAR-100 and Tiny ImageNet, the embedding-size effect rather than the proposed mechanism explains the result. A second check would verify whether the mutual information curves $I(X,C)$ and $I(C,Y)$ in Fig. 5 remain monotonic for that baseline; if they do, the information-retention claim is not specific to LCBM.

Watch

Extended reading notes

Core claim

On the paper's own terms, the discovery is that modelling each unsupervised concept as a Bernoulli random variable with an associated 128-dimensional embedding, rather than a single scalar activation, lets a concept bottleneck carry enough information to nearly match black-box accuracy without losing local linear interpretability. The model is trained with a variational ELBO whose three terms correspond to representativity (input reconstruction), completeness (task classification), and alignment (KL to a prior), and the final class prediction is a linear combination of concept scores with weights produced from the embeddings. The reported evidence includes accuracy gaps below 1-2% on CIFAR-10, CIFAR-100, Tiny ImageNet, Skin Lesions, and CUB-200, higher mutual information with both input and label during training, lower reconstruction error, higher concept-F1 and concept-alignment scores against human annotations, and a user study where participants more often identify intruder and completing images for LCBM concepts than for the BotCL baseline.

Load-bearing premise

The claim that LCBM's architecture, rather than its larger concept representation, drives the accuracy gain is the load-bearing assumption: baselines use a single scalar per concept while LCBM uses a 128-dimensional embedding, and the Appendix C ablation only changes embedding size without ever matching baselines in capacity.

Editorial extensions

If this is right

  • Unsupervised concept models become usable in domains where concept labels and reliable language-model priors are unavailable, since LCBM derives concepts from data alone while staying within roughly 1-2% of black-box accuracy.
  • With a small fixed number of concepts (10-30), LCBM scales to 100-200 class datasets, suggesting concept bottlenecks do not need class-scaled concept sets.
  • Prediction remains locally interpretable as a weighted sum of concept activations, and negative concept interventions change predictions, so users can probe and counterfactually edit decisions.
  • The higher concept-F1 and CAS scores imply the discovered concepts can double as a dataset annotation source, reducing manual labelling effort.

Reading between the lines

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

  • The reported gains may be partly a capacity effect rather than a property of the embedding mechanism: the baselines pass one scalar per concept, while LCBM passes 128 dimensions, and the Appendix C ablation varies embedding size but does not give a capacity-matched baseline.
  • Because the classifier weights are sample-dependent, the 'linear combination' is local, not global; a user cannot read a single fixed rule from the model, only per-image explanations, and whether that satisfies regulated interpretability requirements is an open question the paper does not settle.
  • The authors themselves note the CNN decoder limits reconstruction quality and that manual inspection of concepts is still required, so the representativity and interpretability gains are bounded by that component.
  • If concept embeddings are this informative, the same representation could be reused for auxiliary tasks such as concept-based retrieval, dataset audit, or generating natural-language concept descriptions, but those applications are not tested here.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 4 minor

Summary. The manuscript introduces the Learnable Concept-Based Model (LCBM), an unsupervised concept-based image classifier whose latent concepts are Bernoulli random variables augmented with per-concept continuous embeddings in R^d. The model is trained with an ELBO objective combining a reconstruction term, a classification term, and a KL term with a batch-regularized Bernoulli prior; predictions are made as a local linear combination of concept scores, with sample-dependent weights predicted from the embeddings. The authors evaluate LCBM on seven image datasets against SENN, BotCL (two variants), ProtoPNet, LF-CBM, and an end-to-end black-box baseline, using task accuracy, information-plane curves, reconstruction MSE, concept-F1/CAS alignment metrics, a 72-participant user study, and negative intervention experiments. The central claims are that LCBM surpasses existing unsupervised concept-based models in generalization, nearly matches black-box accuracy with a gap "always less than 1-2%," and yields more interpretable concepts.

Significance. If the claims hold, the paper makes a useful contribution to interpretable machine learning: it proposes a concrete mechanism—concept embeddings combined with Bernoulli scores—for improving the accuracy of unsupervised concept bottlenecks while retaining linear, local interpretability. The release of code, the use of multiple datasets and baselines, the standard variational derivation, and the user study with 72 participants are strengths that make the empirical claims partly reproducible. However, the headline generalization claims are overstated in their current form, and the attribution of the accuracy gain to the embedding mechanism is not yet substantiated because no capacity-matched baseline is presented. The contribution is potentially significant, but the present version requires scoping and additional experiments before the central claims can be accepted as stated.

major comments (3)
  1. [Section 4.2, Fig. 4, Table 4 (Appendix F)] The claim in Section 4.2 that "the generalization loss is always less than 1-2%" relative to the E2E black-box model is contradicted by the paper's own Table 4. On CUB-200 with ResNet-18, LCBM achieves 60.52±1.01% accuracy while E2E achieves 67.96±0.34%, a gap of 7.44 percentage points—more than three times the stated bound. In addition, the claim that LCBM "surpasses existing unsupervised concept-based models" fails on MNIST Even/Odd, where LCBM at 97.25±0.97% trails BotCL (contr) at 98.01±0.29% and LF-CBM at 97.68±0.32%. The qualitative statement in the text that "only ... a few methods perform better, by a few decimals" understates these gaps. The authors should scope the claim (e.g., to "most benchmarks") or explicitly report and discuss the CUB-200 gap; as written, the unqualified statements in the abstract and Section 4.2 are not supported by the reported numbers.
  2. [Section 3.2, Appendix C] The paper attributes LCBM's accuracy improvement to "unsupervised concept embeddings," but the comparison baselines (SENN, BotCL, ProtoPNet, LF-CBM) use a single scalar value per concept, whereas LCBM uses a 128-dimensional vector per concept. This is a substantial difference in bottleneck capacity and parameter count, yet no capacity-matched baseline is provided: Appendix C only varies the embedding dimension within LCBM itself (Fig. 10), and does not compare against baselines given an equivalent per-concept parameter budget. Without such a comparison, the claim that the gain comes from the embedding mechanism rather than from the larger representation capacity is not established. A concrete test would be to train the baselines with an increased per-concept capacity (or to ablate LCBM to a single-scalar concept representation while keeping the same training procedure) and report whether the accuracy advantage persists.
  3. [Section 4.3, Fig. 5] The information-plane analysis in Fig. 5 is central to the claim that LCBM "retains more information regarding both the input and the output," but the mutual information estimator is not specified. For continuous concept embeddings and discrete/continuous inputs, I(X,C) and I(C,Y) are generally intractable, and the reported values depend heavily on the estimator, the discretization, and the number of samples. Without specifying the estimator (e.g., k-NN, histogram, variational bound) and its hyperparameters, these curves cannot be reproduced or compared across models with different representation types. The authors should state the estimator and settings, or otherwise present an alternative quantitative justification for the information-retention claim.
minor comments (4)
  1. [Appendix A, Eq. (13)] The label "Aignement" in Eq. (13) is a typo for "Alignment."
  2. [Figures 17, 18, 24] The figure captions contain typos: "extracted form" should be "extracted from," and "MNIST-Evan/Odd" should be "MNIST Even/Odd."
  3. [Section 4.1] The description of the information-plane metric refers to "the evolution of the mutual information ... as the training epoch increases," but Fig. 5 appears to show only the final information-plane position per model. If the figure shows trajectories, the axis and legend should be clarified; if it shows only endpoint values, the text should be updated accordingly.
  4. [Table 3 (Appendix B)] The ViT results are presented in an appendix without a corresponding discussion of whether the observed patterns (e.g., LCBM trailing BotCL on MNIST Even/Odd by 0.3%) affect the generalizability of the main claims. A sentence relating these results to the abstract claims would help.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity: the ELBO derivation and architecture are self-contained, and the main claims are empirical benchmarks against external baselines. The overbroad 'always less than 1-2%' gap claim is a correctness/scoping issue, not circularity.

full rationale

The derivation chain is self-contained. The ELBO (Eq. 3) is obtained in Appendix A from the standard KL decomposition log p(x,y) = ELBO + KL(q||p), followed by the conditional-independence assumption p(x,y|c)=p(x|c)p(y|c); none of these steps presupposes LCBM's accuracy or interpretability claims. The Bernoulli concept scores, prototype dot-products, embeddings, and linear classifier are architectural choices, and hyperparameters (alpha=0.2, d=128) are selected by validation or ablation, not fitted to quantities later reported as predictions. The main claims ('surpasses existing unsupervised concept-based models', 'generalization loss is always less than 1-2%') are empirical benchmark statements evaluated against external baselines (SENN, BotCL, ProtoPNet, LF-CBM, E2E) and therefore are not definitional. Self-citations ([8] for CAS, [28] for the concept-XAI framing) define metrics or terminology and are not load-bearing for the core derivation. The user study relies on the model's own concept dictionaries, which introduces selection bias, but it is an evaluation protocol rather than a circular equation-level step. I note one non-circular weakness: Table 4 shows LCBM trails E2E by 7.44 points on CUB-200, contradicting the unqualified 'always less than 1-2%' claim; this is an accuracy/scoping issue, not circularity.

Assumptions & free parameters 3 free parameters · 6 assumptions · 0 invented entities

The central claim rests on the conditional-independence factorization of the joint distribution, the sufficiency of the frozen backbone features, the binary atomicity assumption for concepts, and three hand-chosen hyperparameters (concept counts, embedding size, and prior alpha). No new external physical or semantic entities are postulated.

free parameters (3)
  • Number of concepts per dataset = CIFAR-10: 15, CIFAR-100: 20, Tiny ImageNet: 30, Skin Lesions: 14, CUB-200: 112 (MNIST: 10)
    Hand-chosen per dataset; directly controls the size of the concept bottleneck and is a key determinant of the accuracy-interpretability trade-off.
  • Concept embedding dimensionality d = 128
    Chosen after an ablation (Appendix C) showing accuracy plateaus around d=128; a free hyperparameter not derived from theory.
  • Bernoulli prior probability alpha = 0.2
    Hyperparameter for the BatchPrior regularization; manually set; controls expected activation probability per concept per sample.
assumptions (6)
  • domain assumption Conditional independence p(x,y|c)=p(x|c)p(y|c)
    Invoked in Appendix A, Eq. (13), to factor the joint distribution; assumes the latent concepts mediate all shared information between input and label.
  • domain assumption The data generating process p(x|c)p(y|c)p(c) is adequate for image classification
    Section 3.1, Eq. (2); the model assumes images and labels are conditionally independent given the concepts, and that the concepts capture all task-relevant information.
  • domain assumption Frozen backbone g provides sufficient features for both reconstruction and classification
    Section 3.2; concept embeddings are computed as h_j(g(x)) with g fixed; if g lacks task-relevant features, the bottleneck cannot be complete.
  • domain assumption Concepts are atomic binary units (Bernoulli)
    Section 3.2; discreteness is assumed to aid interpretability and intervention, but restricts the representational capacity of scores.
  • domain assumption Mutual information estimates in Fig. 5 are valid
    Section 4.3; the paper does not specify the estimator, so the information plane results rest on an unstated measurement assumption.
  • standard math Reparameterized Bernoulli sampling exists for discrete distributions
    Section 3.2, Eq. (4); relies on the Concrete/Gumbel-softmax reparameterization trick [22].

how reviews work

0 comments
Cite this review

Pith. "Pith review of Towards Better Generalization and Interpretability in Unsupervised Concept-Based Models." pith.science (2026). https://pith.science/paper/Z74BDFDW

@misc{pith2026250602092,
  author       = {Pith},
  title        = {Pith review of: Towards Better Generalization and Interpretability in Unsupervised Concept-Based Models},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/Z74BDFDW}},
  note         = {Machine review of arXiv:2506.02092}
}
read the original abstract

To increase the trustworthiness of deep neural networks, it is critical to improve the understanding of how they make decisions. This paper introduces a novel unsupervised concept-based model for image classification, named Learnable Concept-Based Model (LCBM) which models concepts as random variables within a Bernoulli latent space. Unlike traditional methods that either require extensive human supervision or suffer from limited scalability, our approach employs a reduced number of concepts without sacrificing performance. We demonstrate that LCBM surpasses existing unsupervised concept-based models in generalization capability and nearly matches the performance of black-box models. The proposed concept representation enhances information retention and aligns more closely with human understanding. A user study demonstrates the discovered concepts are also more intuitive for humans to interpret. Finally, despite the use of concept embeddings, we maintain model interpretability by means of a local linear combination of concepts.

Figures

Figures reproduced from arXiv: 2506.02092 by the authors.

Figure 1
Figure 1. Learnable Concept-Based Model (LCBM) learns a dictionary of un￾supervised concepts. Unlike black-box models, LCBM classifies images inter￾pretably using these concepts. Here, the image is correctly classified as Dark-eyed junco by leveraging con￾cepts C48 (eyes/beak), C47 (wings), and C71 (trunk/tree). Notably, C71, while present, is less relevant to bird species classification. CBMs can be created in a su￾pervised … view at source ↗
Figure 2
Figure 2. Probabilistic Graphical Model. Solid arrows represent the data generating [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. LCBM schema. The concept encoder q(c | x) provides the probability for each learnt concept cˆj and the associated embeddings cj . Both concept scores and embeddings are used to predict the output class p(y | c) and to reconstruct the input p(x | c). Optimization problem. LCBMs are trained to optimize the log-likelihood of tuples (x, y). Following a variational inference approach, we optimize the ev￾idence lower boun… view at source ↗
Figures from the paper (17 more)
Figure 4
Figure 4. Figure 4: Comparison of the generalization performance across the evaluated [PITH_FULL_IMAGE:figures/full_fig_p008_4.png]
Figure 5
Figure 5. Figure 5: Information Plane for the different models in terms of Mutual information [PITH_FULL_IMAGE:figures/full_fig_p009_5.png]
Figure 6
Figure 6. Figure 6: Tiny-Imagenet dictionary produced by LCBM. Each column of images [PITH_FULL_IMAGE:figures/full_fig_p011_6.png]
Figure 8
Figure 8. Figure 8: Negative interventions: percent￾age accuracy loss when increasing the intervention probability. The higher the accuracy loss, the higher the sensi￾tivity of the model to human interven￾tions. Results for all datasets can be found in Appendix I. concept, as proposed in …
Figure 9
Figure 9. Figure 9: Example of interpretable prediction on different datasets. We provide [PITH_FULL_IMAGE:figures/full_fig_p013_9.png]
Figure 10
Figure 10. Figure 10: Effect of concept embedding size on task performance. As the embedding [PITH_FULL_IMAGE:figures/full_fig_p019_10.png]
Figure 11
Figure 11. Figure 11: Characterization of the users involved in the user study. [PITH_FULL_IMAGE:figures/full_fig_p021_11.png]
Figure 12
Figure 12. Figure 12: The figure presents an example for each type of question displayed to [PITH_FULL_IMAGE:figures/full_fig_p022_12.png]
Figure 13
Figure 13. Figure 13: CIFAR10 dictionary of concepts. as concept C12, which groups images of people, while others are more general, such as concept C13, which represents ‘landscapes with trees’. Among these types of concepts is concept C2, which group objects with a white background. Many …
Figure 14
Figure 14. Figure 14: CIFAR-100 dictionary of concepts. The MNIST dictionary shown in [PITH_FULL_IMAGE:figures/full_fig_p024_14.png]
Figure 15
Figure 15. Figure 15: MNIST Even/Odd dictionary of concepts. A similar phenomenon occurs with the concept dictionary of MNIST Addi￾tion, shown in [PITH_FULL_IMAGE:figures/full_fig_p025_15.png]
Figure 16
Figure 16. Figure 16: MNIST Addition dictionary of concepts. (representing architectures/constructions) and concept C10 (representing the sky). Grad-CAMs reveal that C24 focuses on a specific section of the bridge, while C2 highlights the sky. On the other hand, [PITH_FULL_IMAGE:figures/f…
Figure 17
Figure 17. Figure 17: Example belonging to class "odd" extracted form the MNIST￾Evan/Odd dataset. Input image 6 C6 Explanation ‘even’ 4 0 C7 C8 C9 Importance [PITH_FULL_IMAGE:figures/full_fig_p027_17.png]
Figure 19
Figure 19. Figure 19: Example belonging to class "viaduct" extracted form the Tiny Imagenet dataset. Input image 10 C11 Explanation ‘kimono’ 5 0 C5 C13C22 Importance [PITH_FULL_IMAGE:figures/full_fig_p027_19.png]
Figure 21
Figure 21. Figure 21: Example belonging to class "truck" extracted form the CIFAR￾10 dataset. Input image 4 C0 Explanation ‘cat’ 2 0 C5 C12 C14 Importance [PITH_FULL_IMAGE:figures/full_fig_p028_21.png]
Figure 23
Figure 23. Figure 23: Example belonging to class "turtle" extracted form the CIFAR￾100 dataset. Input image 2 C2 Explanation ‘rat’ 1 0 C6 C7 C15 Importance [PITH_FULL_IMAGE:figures/full_fig_p028_23.png]
Figure 25
Figure 25. Figure 25: Impact of negative interventions across different datasets. The figure [PITH_FULL_IMAGE:figures/full_fig_p028_25.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

46 extracted references · 26 canonical work pages

  1. [1]

    Nature Machine Intelligence5(9), 1006– 1019 (2023)

    Achtibat, R., Dreyer, M., Eisenbraun, I., Bosse, S., Wiegand, T., Samek, W., Lapuschkin, S.: From attribution maps to human-understandable explanations through concept relevance propagation. Nature Machine Intelligence5(9), 1006– 1019 (2023)

    2023

  2. [2]

    IEEE access6, 52138–52160 (2018)

    Adadi, A., Berrada, M.: Peeking inside the black-box: a survey on explainable artificial intelligence (xai). IEEE access6, 52138–52160 (2018)

    2018

  3. [3]

    Advances in neural information processing systems 31(2018)

    Alvarez Melis, D., Jaakkola, T.: Towards robust interpretability with self- explaining neural networks. Advances in neural information processing systems 31(2018)

    2018

  4. [4]

    In: Proceedings of the AAAI Confer- ence on Artificial Intelligence

    Barbiero, P., Ciravegna, G., Giannini, F., Lió, P., Gori, M., Melacci, S.: Entropy- based logic explanations of neural networks. In: Proceedings of the AAAI Confer- ence on Artificial Intelligence. vol. 36, pp. 6046–6054 (2022)

    work page 2022

  5. [5]

    IEEE transactions on pattern analysis and machine intelligence 35(8), 1798–1828 (2013)

    Bengio, Y., Courville, A., Vincent, P.: Representation learning: A review and new perspectives. IEEE transactions on pattern analysis and machine intelligence 35(8), 1798–1828 (2013)

    2013

  6. [6]

    Advances in neural information processing systems32(2019)

    Chen, C., Li, O., Tao, D., Barnett, A., Rudin, C., Su, J.K.: This looks like that: deep learning for interpretable image recognition. Advances in neural information processing systems32(2019)

    2019

  7. [7]

    Artificial Intelligence314, 103822 (2023)

    Ciravegna, G., Barbiero, P., Giannini, F., Gori, M., Lió, P., Maggini, M., Melacci, S.: Logic explained networks. Artificial Intelligence314, 103822 (2023)

    work page 2023

  8. [8]

    In: Koyejo, S., Mohamed, S., Agarwal, A., Belgrave, D., Cho, K., Oh, A

    Espinosa Zarlenga, M., Barbiero, P., Ciravegna, G., Marra, G., Giannini, F., Dili- genti, M., Shams, Z., Precioso, F., Melacci, S., Weller, A., Lió, P., Jamnik, M.: Con- cept embedding models: Beyond the accuracy-explainability trade-off. In: Koyejo, S., Mohamed, S., Agarwal, A., Belgrave, D., Cho, K., Oh, A. (eds.) Advances in Neural Information Processi...

    work page 2022

Show all 46 references
  1. [9]

    Advances in Neural Information Processing Systems36(2024)

    Espinosa Zarlenga, M., Collins, K., Dvijotham, K., Weller, A., Shams, Z., Jam- nik, M.: Learning to receive help: Intervention-aware concept embedding models. Advances in Neural Information Processing Systems36(2024)

    2024

  2. [10]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition

    Fel, T., Picard, A., Bethune, L., Boissin, T., Vigouroux, D., Colin, J., Cadène, R., Serre, T.: Craft: Concept recursive activation factorization for explainability. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 2711–2721 (2023)

    2023

  3. [11]

    Advances in neural information processing systems32(2019)

    Ghorbani, A., Wexler, J., Zou, J.Y., Kim, B.: Towards automatic concept-based explanations. Advances in neural information processing systems32(2019)

    2019

  4. [12]

    ACM computing surveys (CSUR)51(5), 1–42 (2018)

    Guidotti, R., Monreale, A., Ruggieri, S., Turini, F., Giannotti, F., Pedreschi, D.: A survey of methods for explaining black box models. ACM computing surveys (CSUR)51(5), 1–42 (2018)

    2018

  5. [13]

    Hase, P., Chen, C., Li, O., Rudin, C.: Interpretable image recognition with hierar- chicalprototypes.In:ProceedingsoftheAAAIConferenceonHumanComputation and Crowdsourcing. vol. 7, pp. 32–40 (2019)

    2019

  6. [14]

    In: International conference on machine learning

    Kim, B., Wattenberg, M., Gilmer, J., Cai, C., Wexler, J., Viegas, F., et al.: Inter- pretabilitybeyondfeatureattribution:Quantitativetestingwithconceptactivation vectors (tcav). In: International conference on machine learning. pp. 2668–2677. PMLR (2018)

    2018

  7. [15]

    In: Proceedings of the 40th International Conference on Machine Learning

    Kim, E., Jung, D., Park, S., Kim, S., Yoon, S.: Probabilistic concept bottleneck models. In: Proceedings of the 40th International Conference on Machine Learning. ICML’23, JMLR.org (2023)

    2023

  8. [16]

    In: International conference on machine learning

    Koh, P.W., Nguyen, T., Tang, Y.S., Mussmann, S., Pierson, E., Kim, B., Liang, P.: Concept bottleneck models. In: International conference on machine learning. pp. 5338–5348. PMLR (2020)

    2020

  9. [17]

    Krizhevsky, A., et al.: Learning multiple layers of features from tiny images (2009)

    2009

  10. [18]

    Naval research logistics quarterly2(1-2), 83–97 (1955)

    Kuhn, H.W.: The hungarian method for the assignment problem. Naval research logistics quarterly2(1-2), 83–97 (1955)

    1955

  11. [19]

    In: Proceedings of the 2019 AAAI/ACM Con- ference on AI, Ethics, and Society

    Lakkaraju, H., Kamar, E., Caruana, R., Leskovec, J.: Faithful and customizable explanations of black box models. In: Proceedings of the 2019 AAAI/ACM Con- ference on AI, Ethics, and Society. pp. 131–138 (2019)

    2019

  12. [20]

    Proceedings of the IEEE86(11), 2278–2324 (1998).https: //doi.org/10.1109/5.726791

    Lecun, Y., Bottou, L., Bengio, Y., Haffner, P.: Gradient-based learning applied to document recognition. Proceedings of the IEEE86(11), 2278–2324 (1998).https: //doi.org/10.1109/5.726791

    1998 doi

  13. [21]

    In: Proceedings of the AAAI Conference on Artificial Intelligence

    Li,O.,Liu,H.,Chen,C.,Rudin,C.:Deeplearningforcase-basedreasoningthrough prototypes: A neural network that explains its predictions. In: Proceedings of the AAAI Conference on Artificial Intelligence. vol. 32 (2018)

    2018

  14. [22]

    In: International Conference on Learning Representations (2022)

    Maddison, C.J., Mnih, A., Teh, Y.W.: The concrete distribution: A continuous relaxation of discrete random variables. In: International Conference on Learning Representations (2022)

    2022

  15. [23]

    Advances in Neural Information Processing Systems35, 21212–21227 (2022)

    Marconato, E., Passerini, A., Teso, S.: Glancenets: Interpretable, leak-proof concept-based models. Advances in Neural Information Processing Systems35, 21212–21227 (2022)

    2022

  16. [24]

    Psychological review63(2), 81 (1956)

    Miller, G.A.: The magical number seven, plus or minus two: Some limits on our capacity for processing information. Psychological review63(2), 81 (1956)

    1956

  17. [25]

    Advances in Neural Information Processing Sys- tems35, 4667–4679 (2022)

    Misino, E., Marra, G., Sansone, E.: Vael: Bridging variational autoencoders and probabilistic logic programming. Advances in Neural Information Processing Sys- tems35, 4667–4679 (2022)

    2022

  18. [26]

    In: The Eleventh International Conference on Learning Representations (2023),https://openreview.net/forum?id=FlCg47MNvBA 16 F

    Oikarinen, T., Das, S., Nguyen, L.M., Weng, T.W.: Label-free concept bottleneck models. In: The Eleventh International Conference on Learning Representations (2023),https://openreview.net/forum?id=FlCg47MNvBA 16 F. De Santis et al

    2023

  19. [27]

    In: Proceedings of the 2023 ACM Conference on Fairness, Accountability, and Transparency

    Panigutti, C., Hamon, R., Hupont, I., Fernandez Llorca, D., Fano Yela, D., Jun- klewitz, H., Scalzo, S., Mazzini, G., Sanchez, I., Soler Garrido, J., et al.: The role of explainable ai in the context of the ai act. In: Proceedings of the 2023 ACM Conference on Fairness, Accoun...

    2023

  20. [28]

    arXiv preprint arXiv:2312.12936 (2023)

    Poeta, E., Ciravegna, G., Pastor, E., Cerquitelli, T., Baralis, E.: Concept-based explainable artificial intelligence: A survey. arXiv preprint arXiv:2312.12936 (2023)

    2023

  21. [29]

    In: International conference on machine learning

    Radford, A., Kim, J.W., Hallacy, C., Ramesh, A., Goh, G., Agarwal, S., Sastry, G., Askell, A., Mishkin, P., Clark, J., et al.: Learning transferable visual models from natural language supervision. In: International conference on machine learning. pp. 8748–8763. PMLR (2021)

    2021

  22. [30]

    why should i trust you?

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

    2016

  23. [31]

    Nature machine intelligence1(5), 206–215 (2019)

    Rudin, C.: Stop explaining black box machine learning models for high stakes decisions and use interpretable models instead. Nature machine intelligence1(5), 206–215 (2019)

    2019

  24. [32]

    IEEE Access10, 41758–41765 (2022)

    Sawada, Y., Nakamura, K.: Concept bottleneck model with additional unsuper- vised concepts. IEEE Access10, 41758–41765 (2022)

    2022

  25. [33]

    In: Proceedings of the IEEE international conference on computer vision

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

    2017

  26. [34]

    arXiv preprint arXiv:1703.00810 (2017)

    Shwartz-Ziv, R., Tishby, N.: Opening the black box of deep neural networks via information. arXiv preprint arXiv:1703.00810 (2017)

    2017 arXiv

  27. [35]

    Srivastava, D., Yan, G., Weng, L.: Vlg-cbm: Training concept bottleneck models withvision-languageguidance.AdvancesinNeuralInformationProcessingSystems 37, 79057–79094 (2024)

    2024

  28. [36]

    arXiv preprint physics/0004057 (2000)

    Tishby, N., Pereira, F.C., Bialek, W.: The information bottleneck method. arXiv preprint physics/0004057 (2000)

    2000 arXiv

  29. [37]

    Scientific data5(1), 1–9 (2018)

    Tschandl, P., Rosendahl, C., Kittler, H.: The ham10000 dataset, a large collection of multi-source dermatoscopic images of common pigmented skin lesions. Scientific data5(1), 1–9 (2018)

    2018

  30. [38]

    Computer Law Review International22(4), 97–112 (2021)

    Veale, M., Zuiderveen Borgesius, F.: Demystifying the draft eu artificial intelli- gence act—analysing the good, the bad, and the unclear elements of the proposed approach. Computer Law Review International22(4), 97–112 (2021)

    2021

  31. [39]

    Wah,C.,Branson,S.,Welinder,P.,Perona,P.,Belongie,S.:Thecaltech-ucsdbirds- 200-2011 dataset. Tech. Rep. CNS-TR-2011-001, California Institute of Technology (2011)

    2011

  32. [40]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition

    Wang, B., Li, L., Nakashima, Y., Nagahara, H.: Learning bottleneck concepts in image classification. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 10962–10971 (2023)

    2023

  33. [41]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition

    Yang, Y., Panagopoulou, A., Zhou, S., Jin, D., Callison-Burch, C., Yatskar, M.: Language in a bottle: Language model guided concept bottlenecks for interpretable image classification. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 19...

    2023

  34. [42]

    CS 231N2(5), 8 (2015)

    Yao, L., Miller, J.: Tiny imagenet classification with convolutional neural networks. CS 231N2(5), 8 (2015)

    2015

  35. [43]

    Advances in neural information processing systems33, 20554–20565 (2020) Towards Better Generalization and Interpretability in Unsupervised CBMs 17

    Yeh, C.K., Kim, B., Arik, S., Li, C.L., Pfister, T., Ravikumar, P.: On completeness- aware concept-based explanations in deep neural networks. Advances in neural information processing systems33, 20554–20565 (2020) Towards Better Generalization and Interpretability in Unsuperv...

    2020

  36. [44]

    In: ICLR 2022 Workshop on PAIR^2Struct: Privacy, Accountability, Interpretability, Ro- bustness, Reasoning on Structured Data (2022),https://openreview.net/forum? id=HAMeOIRD_g9

    Yuksekgonul, M., Wang, M., Zou, J.: Post-hoc concept bottleneck models. In: ICLR 2022 Workshop on PAIR^2Struct: Privacy, Accountability, Interpretability, Ro- bustness, Reasoning on Structured Data (2022),https://openreview.net/forum? id=HAMeOIRD_g9

    2022

  37. [45]

    In: Computer Vision–ECCV 2014: 13th European Conference, Zurich, Switzerland, September 6-12, 2014, Proceedings, Part I 13

    Zeiler, M.D., Fergus, R.: Visualizing and understanding convolutional networks. In: Computer Vision–ECCV 2014: 13th European Conference, Zurich, Switzerland, September 6-12, 2014, Proceedings, Part I 13. pp. 818–833. Springer (2014)

    2014

  38. [46]

    food," "pizza,

    Zhang, Q., Wu, Y.N., Zhu, S.C.: Interpretable convolutional neural networks. In: Proceedings of the IEEE conference on computer vision and pattern recognition. pp. 8827–8836 (2018) 18 F. De Santis et al. A Derivation details Sincecis unknown, we estimate it leveraging the obse...

    2018

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.