Pith. sign in

REVIEW 5 major objections 6 minor 43 references

PXGen: A Post-hoc Explainable Method for Generative Models

T0 review · 5 major / 6 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read A post-hoc method identifies the most representative training samples of an encoder-decoder generative model without accessing the training phase, using only reconstruction error and latent-space divergence.

desk verdict PXGen is a clear, sensible recipe for post-hoc explanation of VAEs, but the representativeness claim is validated circularly and without error bars, so the headline comparison to VAE-TracIn does not hold as presented. read the letter →

arxiv 2501.11827 v1 pith:OI24BMMH submitted 2025-01-21 cs.LG cs.AI

Yen-Lung Huang , Ming-Hsi Weng , Hao-Tsung Yang This is my paper · ORCID
classification cs.LGcs.AI
keywords explainableAIgenerativemodelspost-hocexplanationvariationalautoencoderrepresentativesamplesinfluenceanalysismodeldelusionanchorset
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper proposes PXGen, a post-hoc explainable framework for generative models that works without access to training data or training dynamics. Given a model and a user-chosen anchor set, PXGen scores each anchor by two criteria: how far the model's encoded distribution for that sample is from the prior, and how well the model reconstructs the sample. A threshold derived from generated samples splits anchors into four groups, exposing two model behaviors: 'model delusion' (the model encodes confidently but reconstructs incorrectly) and 'aligned conception' (low latent affinity but accurate reconstruction). The paper's main empirical claim is that the high-affinity group contains the most representative training samples: models retrained only on that group generate images closer to the original model than random removal or VAE-TracIn-based removal, measured by FID. This gives a tractable, no-training-access route to auditing and curating generative models.

What carries the argument

The load-bearing mechanism is the anchor set plus an intrinsic and an extrinsic criterion. The intrinsic criterion is the KLD between the distribution the encoder assigns to an anchor and the model's prior $N(0,I)$; the extrinsic criterion is the reconstruction error between an anchor and its decoded version (MSE for MNIST, FID for CIFAR-10). Thresholds are computed as the average of the per-iteration maximum values over generated images, making the split depend on the model's typical outputs rather than on the anchor set. The high/low split yields the four groups HIHE, HILE, LIHE, and LILE, and $k$-dispersion or $k$-center then selects representative anchors from a group for visualization.

What would settle it

Retrain a small VAE many times, each time deleting exactly one training sample, and measure the FID shift of the generated distribution relative to the full model; then rank samples by PXGen's HIHE anchor values and check whether the top-ranked samples are enriched in the largest per-sample FID shifts. If the enrichment is no better than random selection, the central claim that HIHE identifies the most representative training samples is undermined.

Watch

Extended reading notes

Core claim

On the paper's own terms, the discovery is that a black-box encoder-decoder generative model can be explained by measuring each candidate sample against two scalar criteria—KLD between the encoded latent distribution and the prior $N(0,I)$, and reconstruction error (MSE for simple images, FID for complex ones)—and then classifying samples as high or low on each axis. The four resulting groups carry distinct behavioral meanings: HIHE samples are what the model genuinely knows; HILE samples expose 'model delusion,' where the model believes it recognizes a concept but decodes it into something else; LIHE samples exhibit 'aligned conception,' where the model reconstructs them accurately despite low latent affinity; LILE samples are furthest from the model. The paper further claims that the HIHE group identifies the most representative training samples, validated by retraining experiments in which the FID distance to the original model is smallest when only HIHE data is kept, compared with random removal and with VAE-TracIn's influence scores.

Load-bearing premise

The load-bearing premise is that a sample's importance to a model is faithfully summarized by two numbers—how far the model's encoded distribution for that sample sits from its prior, and how well the model reconstructs that sample—and that the boundary between 'high' and 'low' affinity is correctly placed at the average of the worst values achieved by the model's own generated samples; if that boundary is misplaced, the four groups and the representative-sample conclusion collapse.

Editorial extensions

If this is right

  • A deployed generative model whose training pipeline is proprietary can still be audited post hoc: PXGen points to concrete samples that most shape what the model generates, with no access to gradients, checkpoints, or training data.
  • The HIHE-versus-rest split gives a simple data-pruning recipe: keeping the high-affinity samples preserves a model's behavior, while removing them shifts behavior maximally.
  • The 'model delusion' and 'aligned conception' groupings turn a trained generator's latent behavior into inspectable categories, which could support debugging of learned biases and spurious concepts.
  • Because anchors need not be training data, users can probe a model with their own samples and see which concepts the model has internalized versus which it only reconstructs by accident.
  • PXGen's $O(n^2)$ cost per criterion makes it applicable at deployment time even for models whose training runs are too large to trace.

Reading between the lines

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

  • If the two scalar criteria really capture representativeness, the same thresholding could be used before training as a data-quality filter, flagging samples that a model is likely to memorize poorly or reconstruct wrongly.
  • The framework's stated generality to GANs and diffusion models is a projection rather than a demonstrated result, since the intrinsic criterion used here is VAE-specific (KLD to a fixed prior); a testable extension would replace it with a latent-space distance defined for those architectures and repeat the removal experiment.
  • The model-delusion group suggests a direct test: an anchor the model encodes confidently but reconstructs wrongly should be the kind of sample that, if added to training data, most improves the model—this is a measurable prediction that the paper does not run.
  • The fuzziness of the 'average maximum' threshold implies that the groups are sensitive to how many generated images are sampled; one could make the method more robust by reporting group membership as a function of the sample size.
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

5 major / 6 minor

Summary. The paper proposes PXGen, a post-hoc explainable framework for generative models. Given a trained model, an anchor set, and user-specified intrinsic/extrinsic criteria (e.g., KLD and MSE), PXGen computes per-anchor feature values, partitions anchors into four groups (HIHE, HILE, LIHE, LILE) using thresholds taken as the average of per-iteration maxima over model-generated samples, and selects characteristic anchors via k-dispersion or k-center algorithms. Demonstrations on a VAE trained on MNIST "0" and a Soft-IntroVAE trained on CIFAR-10 Automobile identify qualitative phenomena called "model delusion" and "aligned conception." The central validation claim (Section 4.4, Figures 7 and 8) is that PXGen can identify the most representative training samples, and that removing them changes the model's generated-output distribution more than random removal or VAE-TracIn-based removal.

Significance. If the central representativeness claim could be supported, PXGen would offer a practical, black-box, training-phase-free XAI method for encoder-decoder generative models, with the attractive properties of customizable multi-criteria input, no training access, and tractable selection algorithms. Strengths of the paper include the explicit post-hoc framing, the flexibility of the anchor/criteria setup, and the attempt to compare against VAE-TracIn. However, the current evaluation is not sufficient to establish the claim: the validation is circular with respect to PXGen's own grouping criteria, the thresholds are ad hoc, no uncertainty quantification is reported, and key experimental constructions are ambiguous. These issues are load-bearing for the paper's central contribution.

major comments (5)
  1. [Section 4.4, Figures 7 and 8] The validation of the representativeness claim is circular. The training data are partitioned into HIHE and other groups using exactly the same KLD+MSE criteria that PXGen uses to define "representative," and the validation then removes or retains those same HIHE samples. Without an external ground truth for influence or representativeness (for example, leave-one-out retraining influence, a held-out likelihood drop, or a metric not used in the grouping), the FID similarity between M-HIHE and the original model only shows that models trained on low-KLD/low-MSE samples resemble a model trained on all samples, which may simply reflect that the original model is dominated by easy samples. The comparison to random removal and VAE-TracIn does not break this circularity because the baseline removals are not matched in terms of the sample-difficulty distribution.
  2. [Section 4.2] The threshold that defines all four anchor groups is ad hoc: it is "the average of the maximum values obtained from multiple iterations" of model-generated samples. The paper does not report the number of iterations, the variance across iterations, or any justification for why the average of maxima is a meaningful high/low cut. Because the grouping changes under monotone transformations of KLD or MSE, the division into HIHE/HILE/LIHE/LILE is not invariant, and every downstream claim about model delusion, aligned conception, and representative samples depends on this arbitrary choice. Either the threshold should be derived from a principled statistical model, or the authors should demonstrate that the reported phenomena are robust across a range of thresholds.
  3. [Figures 7 and 8, Section 4.4] No error bars, multiple seeds, or statistical tests are reported for any of the FID comparisons. The training set for MNIST "0" consists of only 5,923 samples, and the difference between M-HIHE, M-Others, M-Random, and M-TracIn in a single run may be well within retraining noise. The paper should report mean FID over several independent training runs with standard deviations or confidence intervals, and ideally a significance test, before claiming that PXGen 'finds representative training samples well compared with the state-of-the-art.'
  4. [Section 4.4, M-Others construction] The description of M-Others is ambiguous and potentially inconsistent. The sentence "we trained a model M-Others using training data, but removed the training data from the HIHE group in an amount equivalent to that of the other groups, prioritizing the removal of low anchor values" does not clearly specify which samples are retained, which are removed, or how the "amount equivalent to that of the other groups" is computed. Without this detail, the reader cannot determine whether M-Others is a fair baseline with the same training-set size and budget as M-HIHE, and Figure 7's comparison is uninterpretable.
  5. [Section 4.2, 5% analysis] The 5% analysis is not defined precisely: the text says "we focus solely on the subset of anchors with a KLD value within the 5% range among all anchors," but it is not clear whether this means the lowest 5%, the top 5%, or values within 5% of some reference. This selection is also post hoc: it is applied only after observing patterns in the reconstructed images, and it is used to support the model delusion and aligned conception phenomena. The selection rule should be stated explicitly and justified, and the paper should indicate whether the qualitative conclusions change if the 5% is varied.
minor comments (6)
  1. [Throughout] Typographical issues: "three phrases" should be "three phases" (Section 3), "explaination" should be "explanation" (Section 3.1), and "challenginging" should be "challenging" (Section 4.4).
  2. [Section 4.4] The phrase "M-TracIn is trained by removing the same amount of low-help (harmful) training data" is unclear. VAE-TracIn produces influence scores, but "low-help" is not defined; please specify whether low-scoring, negative-scoring, or least representative samples are removed, and how that relates to "harmful."
  3. [Figure 7] The caption says "Left" and "Right" but does not explicitly state that the left panel corresponds to class "0" and the right panel to class "5"; please make this explicit in the caption.
  4. [Section 4.3] The k-dispersion algorithm is described as identifying "the k most distinct individuals," but k-dispersion in the cited literature is an optimization problem with different variants (max-min, max-sum). Please specify which objective is used and how ties are broken.
  5. [Section 3.2] The phrase "Another part involves obtaining information from the model and mapping it to the same multi-dimensional space as the anchors" is vague. Please define what information is obtained (e.g., generated samples, encoder outputs) and how it is mapped.
  6. [Section 3.1] The complexity claim "the calculation time is O(n^2)" is not self-contained. It should specify n (anchor-set size) and clarify whether this covers the criteria computation, the grouping, and the selection algorithms together, or only one of them.

Circularity Check

1 steps flagged · score 4.0 of 10

Representativeness is defined by PXGen's own KLD+MSE score, but the FID and VAE-TracIn validations provide independent evidence, so circularity is partial.

  1. self definitional [Section 4.4, 'Finding Representative Training Samples', pages 10-11]
    "Upon utilizing PXGen, the training dataset can be divided into four groups, and each data point is assigned an anchor value coming from the addition of the KLD and MSE values. According to the discussion in Section 4.2, we believe that the training data within the HIHE group are the most representative... The anchor value indicates the level of representativeness within the group, with lower values being more representative and higher values being less so."

    Representativeness is stipulated to be the summed KLD+MSE anchor value, and HIHE is defined as the set of anchors with low KLD and low MSE under thresholds derived from those same criteria. The paper's central claim that PXGen finds the most representative training samples is therefore, by construction, equivalent to claiming that PXGen finds low-KLD/low-MSE samples. The Section 4.4 experiment then removes exactly the HIHE group and interprets the resulting FID change as confirming representativeness, so the confirmation and the grouping share the same construction. The FID comparison and the VAE-TracIn baseline do provide independent, external checks, which keeps the circularity partial rather than total.

full rationale

The paper's representativeness label in Section 4.4 is not independently grounded: 'anchor value' is defined as the sum of KLD and MSE, and the HIHE group is selected by thresholding those same criteria. That part is self-definitional. However, the validation is not purely internal: FID between models retrained on different subsets is an external observable, and the comparison with VAE-TracIn is an independent baseline, so the central claim has empirical content beyond the definition. No self-citation chains, uniqueness imports, or ansatz-smuggling via citations are present. The absence of random seeds, confidence intervals, and a true exhaustive influence ground truth is a correctness and rigor concern, not circularity by itself.

Assumptions & free parameters 5 free parameters · 4 assumptions · 3 invented entities

The paper's central claims rest on several unvalidated choices: the threshold statistic, the scalar criteria, the additive anchor value, and the qualitative names for groups. None are derived from theory or independently benchmarked.

free parameters (5)
  • KLD threshold = average of per-iteration maximum KLD over generated samples
    Groups anchors into high or low intrinsic affinity; choice determines group membership and all downstream explanations.
  • MSE threshold = average of per-iteration maximum MSE over generated samples
    Groups anchors into high or low extrinsic affinity; choice determines group membership and all downstream explanations.
  • Anchor value weights = 1 for KLD and 1 for MSE
    Representativeness within a group is ranked by the unweighted sum of KLD and MSE; these weights are chosen ad hoc.
  • Extreme percentile = 5%
    The model delusion and aligned conception analyses focus only on anchors within the 5% range of one criterion, a hand-chosen subset.
  • k value = 10
    The number of representative anchors visualized in Figures 5 and 6 is a user-chosen parameter.
assumptions (4)
  • domain assumption The KLD between the encoder's predicted posterior and the prior N(0,I) is a valid measure of how well the model 'knows' an anchor.
    Section 4.1 defines intrinsic affinity this way, but no justification is given that this scalar captures concept recognition.
  • ad hoc to paper The average of per-iteration maximum criterion values over model-generated samples is a valid threshold for splitting high and low affinity.
    Section 4.2 introduces this threshold statistic with no theoretical or empirical comparison to medians, quantiles, or other choices.
  • domain assumption Extrinsic errors such as MSE or FID between an anchor and its reconstruction quantify how much the model 'knows' the anchor.
    Section 4.1 assumes low MSE means the model knows how to generate similar pictures, an unvalidated modeling choice.
  • ad hoc to paper Anchors in the HIHE group are the most representative training samples.
    Section 4.4 states this belief without independent evidence; it is the key premise of the validation experiment.
invented entities (3)
  • Model delusion phenomenon
    purpose: Labels anchors the model encodes confidently but reconstructs poorly, presented as a novel failure mode.
    Qualitative concept introduced in Section 4.2; no metric or benchmark independently validates that this is a real model behavior.
  • Aligned conception phenomenon
    purpose: Labels anchors the model encodes as divergent but reconstructs well, presented as a novel transfer-like behavior.
    Qualitative concept introduced in Section 4.2; no independent evidence beyond the same thresholding procedure.
  • Anchor value
    purpose: Scalar score (KLD plus MSE) used to rank representativeness within a group.
    Introduced in Section 4.4; the additive form and equal weighting are not derived from any theory.

how reviews work

0 comments
Cite this review

Pith. "Pith review of PXGen: A Post-hoc Explainable Method for Generative Models." pith.science (2026). https://pith.science/paper/OI24BMMH

@misc{pith2026250111827,
  author       = {Pith},
  title        = {Pith review of: PXGen: A Post-hoc Explainable Method for Generative Models},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/OI24BMMH}},
  note         = {Machine review of arXiv:2501.11827}
}
read the original abstract

With the rapid growth of generative AI in numerous applications, explainable AI (XAI) plays a crucial role in ensuring the responsible development and deployment of generative AI technologies. XAI has undergone notable advancements and widespread adoption in recent years, reflecting a concerted push to enhance the transparency, interpretability, and credibility of AI systems. Recent research emphasizes that a proficient XAI method should adhere to a set of criteria, primarily focusing on two key areas. Firstly, it should ensure the quality and fluidity of explanations, encompassing aspects like faithfulness, plausibility, completeness, and tailoring to individual needs. Secondly, the design principle of the XAI system or mechanism should cover the following factors such as reliability, resilience, the verifiability of its outputs, and the transparency of its algorithm. However, research in XAI for generative models remains relatively scarce, with little exploration into how such methods can effectively meet these criteria in that domain. In this work, we propose PXGen, a post-hoc explainable method for generative models. Given a model that needs to be explained, PXGen prepares two materials for the explanation, the Anchor set and intrinsic & extrinsic criteria. Those materials are customizable by users according to their purpose and requirements. Via the calculation of each criterion, each anchor has a set of feature values and PXGen provides examplebased explanation methods according to the feature values among all the anchors and illustrated and visualized to the users via tractable algorithms such as k-dispersion or k-center.

Figures

Figures reproduced from arXiv: 2501.11827 by the authors.

Figure 1
Figure 1. The generative model is trained by pictures of the handwriting digit [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. The flowchart of PXGen. In the preparation phase, three items, model, [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. In HILE, the phenomenon of “model delusion” is displayed, where those [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: In LIHE, a phenomenon of “aligned conception” between the handwrit [PITH_FULL_IMAGE:figures/full_fig_p008_4.png]
Figure 5
Figure 5. Figure 5: Utilizing two algorithms to identify representative anchors within HIHE. [PITH_FULL_IMAGE:figures/full_fig_p010_5.png]
Figure 6
Figure 6. Figure 6: Utilizing two algorithms to identify representative anchors within LILE. [PITH_FULL_IMAGE:figures/full_fig_p010_6.png]
Figure 7
Figure 7. Figure 7: Left: Comparing the similarity results generated by comparing the model [PITH_FULL_IMAGE:figures/full_fig_p011_7.png]
Figure 8
Figure 8. Figure 8: We compare our framework with the VAE-TracIn method. by iteratively [PITH_FULL_IMAGE:figures/full_fig_p012_8.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

43 extracted references · 25 canonical work pages

  1. [1]

    Machine learning36, 105–139 (1999)

    Bauer, E., Kohavi, R.: An empirical comparison of voting classification algorithms: Bagging, boosting, and variants. Machine learning36, 105–139 (1999)

    work page 1999

  2. [2]

    JAMA ophthalmology137(3), 258–264 (2019)

    Burlina, P.M., Joshi, N., Pacheco, K.D., Liu, T.A., Bressler, N.M.: Assessment of deep generative models for high-resolution synthetic retinal image generation of age-related macular degeneration. JAMA ophthalmology137(3), 258–264 (2019)

    work page 2019

  3. [3]

    Max-sum diversity via convex programming

    Cevallos, A., Eisenbrand, F., Zenklusen, R.: Max-sum diversity via convex pro- gramming. arXiv preprint arXiv:1511.07077 (2015)

    work page Pith review arXiv 2015

  4. [4]

    In: Proceedings of the Twenty-Eighth Annual ACM-SIAM Symposium on Discrete Algorithms

    Cevallos, A., Eisenbrand, F., Zenklusen, R.: Local search for max-sum diversifi- cation. In: Proceedings of the Twenty-Eighth Annual ACM-SIAM Symposium on Discrete Algorithms. pp. 130–142. SIAM (2017)

    work page 2017

  5. [5]

    Chen, R.: On mentzer’s hardness of the k-center problem on the euclidean plane. (2021)

    work page 2021

  6. [6]

    In: Proceedings of the IEEE/CVF Conference on Com- puter Vision and Pattern Recognition

    Daniel, T., Tamar, A.: Soft-introvae: Analyzing and improving the introspective variational autoencoder. In: Proceedings of the IEEE/CVF Conference on Com- puter Vision and Pattern Recognition. pp. 4391–4400 (2021)

    work page 2021

  7. [7]

    IEEE signal processing magazine29(6), 141–142 (2012)

    Deng, L.: The mnist database of handwritten digit images for machine learning research [best of the web]. IEEE signal processing magazine29(6), 141–142 (2012)

    2012

  8. [8]

    arXiv preprint arXiv:1606.05908 (2016)

    Doersch, C.: Tutorial on variational autoencoders. arXiv preprint arXiv:1606.05908 (2016)

    arXiv 2016

Show all 43 references
  1. [9]

    arXiv preprint arXiv:1802.04208 (2018)

    Donahue, C., McAuley, J., Puckette, M.: Adversarial audio synthesis. arXiv preprint arXiv:1802.04208 (2018)

    2018 arXiv

  2. [10]

    ACM Computing Surveys55(9), 1–33 (2023)

    Dwivedi, R., Dave, D., Naik, H., Singhal, S., Omer, R., Patel, P., Qian, B., Wen, Z., Shah, T., Morgan, G., et al.: Explainable ai (xai): Core ideas, techniques, and solutions. ACM Computing Surveys55(9), 1–33 (2023)

    2023

  3. [11]

    Advances In Neural Information Processing Systems35, 31841–31854 (2022)

    Gao, J., Shen, T., Wang, Z., Chen, W., Yin, K., Li, D., Litany, O., Gojcic, Z., Fidler, S.: Get3d: A generative model of high quality 3d textured shapes learned from images. Advances In Neural Information Processing Systems35, 31841–31854 (2022)

    2022

  4. [12]

    Journal of Artificial Intelligence Research 61, 65–170 (2018)

    Gatt,A.,Krahmer,E.:Surveyofthestateoftheartinnaturallanguagegeneration: Core tasks, applications and evaluation. Journal of Artificial Intelligence Research 61, 65–170 (2018)

    2018

  5. [13]

    Advances in neural infor- mation processing systems27 (2014)

    Goodfellow, I., Pouget-Abadie, J., Mirza, M., Xu, B., Warde-Farley, D., Ozair, S., Courville, A., Bengio, Y.: Generative adversarial nets. Advances in neural infor- mation processing systems27 (2014)

    2014

  6. [14]

    Science robotics4(37), eaay7120 (2019)

    Gunning, D., Stefik, M., Choi, J., Miller, T., Stumpf, S., Yang, G.Z.: Xai—explainable artificial intelligence. Science robotics4(37), eaay7120 (2019)

    2019

  7. [15]

    arXiv preprint arXiv:2012.15781 (2020)

    Guo, H., Rajani, N.F., Hase, P., Bansal, M., Xiong, C.: Fastif: Scalable influ- ence functions for efficient model interpretation and debugging. arXiv preprint arXiv:2012.15781 (2020)

    2020 arXiv

  8. [16]

    Business Insider (2020)

    Hamilton, I.A.: An ai tool which reconstructed a pixelated picture of barack obama to look like a white man perfectly illustrates racial bias in algorithms. Business Insider (2020)

    2020

  9. [17]

    arXiv preprint arXiv:2212.04612 (2022)

    Hammoudeh, Z., Lowd, D.: Training data influence analysis and estimation: A survey. arXiv preprint arXiv:2212.04612 (2022)

    2022 arXiv

  10. [18]

    Operations research letters21(3), 133–137 (1997)

    Hassin, R., Rubinstein, S., Tamir, A.: Approximation algorithms for maximum dispersion. Operations research letters21(3), 133–137 (1997)

    1997

  11. [19]

    Advances in neural information processing systems30 (2017) 14 Huang et al

    Heusel,M.,Ramsauer,H.,Unterthiner,T.,Nessler,B.,Hochreiter,S.:Ganstrained by a two time-scale update rule converge to a local nash equilibrium. Advances in neural information processing systems30 (2017) 14 Huang et al

    2017

  12. [20]

    ICLR (Poster)3 (2017)

    Higgins, I., Matthey, L., Pal, A., Burgess, C.P., Glorot, X., Botvinick, M.M., Mo- hamed, S., Lerchner, A.: beta-vae: Learning basic visual concepts with a con- strained variational framework. ICLR (Poster)3 (2017)

    2017

  13. [21]

    Advances in neural information processing systems33, 6840–6851 (2020)

    Ho, J., Jain, A., Abbeel, P.: Denoising diffusion probabilistic models. Advances in neural information processing systems33, 6840–6851 (2020)

    2020

  14. [22]

    arXiv preprint arXiv:1312.6114 (2013)

    Kingma, D.P., Welling, M.: Auto-encoding variational bayes. arXiv preprint arXiv:1312.6114 (2013)

    2013 arXiv

  15. [23]

    Advances in Neural Information Processing Systems 34, 2400–2412 (2021)

    Kong, Z., Chaudhuri, K.: Understanding instance-based interpretability of vari- ational auto-encoders. Advances in Neural Information Processing Systems 34, 2400–2412 (2021)

    2021

  16. [24]

    The annals of mathe- matical statistics 22(1), 79–86 (1951)

    Kullback, S., Leibler, R.A.: On information and sufficiency. The annals of mathe- matical statistics 22(1), 79–86 (1951)

    1951

  17. [25]

    Advances in Neural Information Processing Systems 32 (2019)

    Lucas, J., Tucker, G., Grosse, R.B., Norouzi, M.: Don’t blame the elbo! a linear vae perspective on posterior collapse. Advances in Neural Information Processing Systems 32 (2019)

    2019

  18. [26]

    Advances in neural information processing systems30 (2017)

    Lundberg, S.M., Lee, S.I.: A unified approach to interpreting model predictions. Advances in neural information processing systems30 (2017)

    2017

  19. [27]

    arXiv preprint arXiv:2208.11970 (2022)

    Luo, C.: Understanding diffusion models: A unified perspective. arXiv preprint arXiv:2208.11970 (2022)

    2022 arXiv

  20. [28]

    Computational Linguistics pp

    Lyu, Q., Apidianaki, M., Callison-Burch, C.: Towards faithful model explanation in nlp: A survey. Computational Linguistics pp. 1–70 (2024)

    2024

  21. [29]

    Unpublished manuscript, March (2016)

    Mentzer, S.G.: Approximability of metric clustering problems. Unpublished manuscript, March (2016)

    2016

  22. [30]

    Advances in Neural Information Processing Systems 33, 19920–19930 (2020)

    Pruthi, G., Liu, F., Kale, S., Sundararajan, M.: Estimating training data influence by tracing gradient descent. Advances in Neural Information Processing Systems 33, 19920–19930 (2020)

    2020

  23. [31]

    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

  24. [32]

    arXiv preprint arXiv:1901.00770 (2019)

    Schneider, J., Handali, J.: Personalized explanation in machine learning: A con- ceptualization. arXiv preprint arXiv:1901.00770 (2019)

    2019 arXiv

  25. [33]

    arXiv preprint arXiv:2404.09554 (2024)

    Schneider, J.: Explainable generative ai (genxai): A survey, conceptualization, and research agenda. arXiv preprint arXiv:2404.09554 (2024)

    2024 arXiv

  26. [34]

    In: Computer Graphics Forum

    Tewari, A., Fried, O., Thies, J., Sitzmann, V., Lombardi, S., Sunkavalli, K., Martin- Brualla, R., Simon, T., Saragih, J., Nießner, M., et al.: State of the art on neural rendering. In: Computer Graphics Forum. vol. 39, pp. 701–727. Wiley Online Li- brary (2020)

    2020

  27. [35]

    In: International Conference on Machine Learning

    Tirinzoni, A., Poiani, R., Restelli, M.: Sequential transfer in reinforcement learning with a generative model. In: International Conference on Machine Learning. pp. 9481–9492. PMLR (2020)

    2020

  28. [36]

    In: Proceedings of the IEEE conference on computer vision and pattern recognition

    Tran, L., Yin, X., Liu, X.: Disentangled representation learning gan for pose- invariant face recognition. In: Proceedings of the IEEE conference on computer vision and pattern recognition. pp. 1415–1424 (2017)

    2017

  29. [37]

    In: Proceedings of the twenty-fifth annual symposium on Computational geometry

    Vattani, A.: K-means requires exponentially many iterations even in the plane. In: Proceedings of the twenty-fifth annual symposium on Computational geometry. pp. 324–332 (2009)

    2009

  30. [38]

    In: International conference on machine learning

    Voynov, A., Babenko, A.: Unsupervised discovery of interpretable directions in the gan latent space. In: International conference on machine learning. pp. 9786–9796. PMLR (2020)

    2020

  31. [39]

    Nature biotechnology38(2), 143–145 (2020) PXGen: A Post-hoc Explainable Method for Generative Models 15

    Walters, W.P., Murcko, M.: Assessing the impact of generative ai on medicinal chemistry. Nature biotechnology38(2), 143–145 (2020) PXGen: A Post-hoc Explainable Method for Generative Models 15

    2020

  32. [40]

    IEEE transactions on image processing 13(4), 600–612 (2004)

    Wang, Z., Bovik, A.C., Sheikh, H.R., Simoncelli, E.P.: Image quality assessment: from error visibility to structural similarity. IEEE transactions on image processing 13(4), 600–612 (2004)

    2004

  33. [41]

    Advances in neural information processing systems 31 (2018)

    Yeh, C.K., Kim, J., Yen, I.E.H., Ravikumar, P.K.: Representer point selection for explaining deep neural networks. Advances in neural information processing systems 31 (2018)

    2018

  34. [42]

    arXiv preprint arXiv:2202.10419 (2022)

    Yin, K., Neubig, G.: Interpreting language models with contrastive explanations. arXiv preprint arXiv:2202.10419 (2022)

    2022 arXiv

  35. [43]

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

    Zhou, J., Joachims, T.: How to explain and justify almost any decision: Potential pitfalls for accountability in ai decision-making. In: Proceedings of the 2023 ACM Conference on Fairness, Accountability, and Transparency. pp. 12–21 (2023)

    2023

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.