Improving Disentangled Representation Learning with the Beta Bernoulli Process

B. Milan Horacek; Cameron Knight; John Sapp; Linwei Wang; Prashnna Kumar Gyawali; Sandesh Ghimire; Zhiyuan Li

REVIEW 4 major objections 4 minor 43 references

Improving Disentangled Representation Learning with the Beta Bernoulli Process

T0 review · 4 major / 4 minor · reviewed 2026-08-14 · deepseek-v4-flash

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

Pith's one-line read The paper claims that a latent density which stays independent but grows in capacity improves VAE disentanglement, and demonstrates it with an Indian Buffet Process prior.

desk verdict Worth reading for the clinical nuisance-factor experiments and the independence-vs-capacity framing, but the ELBO as written is not a valid objective—the Concrete q and Bernoulli prior make the KL infinite, so the stated mechanism is unverified. read the letter →

arxiv 1909.01839 v1 pith:DQEABYJQ submitted 2019-09-03 cs.LG stat.ML

Prashnna Kumar Gyawali , Zhiyuan Li , Cameron Knight , Sandesh Ghimire , B. Milan Horacek , John Sapp , Linwei Wang This is my paper · ORCID

classification cs.LGstat.ML

keywords variationalautoencoderdisentangledrepresentationlearningIndianbuffetprocessbeta-BernoullinonparametriclatentfactormodelposteriormodelingcapacitytotalcorrelationclinicalECG

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 argues that part of what limits disentangled representation learning in variational autoencoders is not only whether latent dimensions are independent, but how much expressive room the latent density has. It claims that when a VAE is regularized toward an independent prior, the prior imposes two separate constraints: independence and a restrictive shape. An independent Gaussian posterior enforces both, and the restrictive shape competes with the reconstruction objective. The paper proposes IBP-VAE, which keeps independence but lets the latent factor model grow with data complexity, and reports that it disentangles more generative factors, including on clinical skin-lesion and ECG data.

What carries the argument

The machinery is the Beta-Bernoulli process, also called the Indian Buffet Process, used as a nonparametric prior over an unbounded binary feature matrix $Z$, combined with a Gaussian magnitude matrix $A$ through the elementwise product $Y = Z \odot A$. This construction keeps latent dimensions independent while allowing a richer marginal density than an isotropic Gaussian. Inference uses a mean-field variational family with a Concrete relaxation for the Bernoulli variables, a Kumaraswamy relaxation for the Beta stick-breaking weights, and a Gaussian $q(A|X)$. The argument runs through the ELBO decomposition into reconstruction, total correlation, dimension-wise KL, and index-code mutual information: richer capacity lowers the dimension-wise KL competition, so independence and reconstruction improve together.

What would settle it

Train IBP-VAE on dSprites with the truncation $K$ set to a small value (for example 10) and again with a large value (for example 200), and measure MIG and total correlation at identical $\beta$. If the small-$K$ model matches the large-$K$ model, the claim that the process grows with data complexity is not load-bearing. A second check: replace the IBP mask with a fixed finite set of independently learned Bernoulli probabilities of the same dimension; if disentanglement remains unchanged, then the nonparametric prior itself is not what drives the improvement.

Watch

Extended reading notes

Core claim

The central claim is that preserving independence among latent factors while increasing the modeling capacity of the posterior density will improve disentangled representation learning in VAEs, especially when generative factors are complex. The paper instantiates this with a latent representation $Y = Z \odot A$, where $Z$ is a binary feature-allocation matrix drawn from a Beta-Bernoulli process (the Indian Buffet Process in the infinite limit) and $A$ is drawn independently from an isotropic Gaussian. Because each data point's active latent features are inferred rather than fixed, the model can in principle grow with the data. On dSprites, IBP-VAE reaches mutual information gap (MIG) 0.5477 at $\beta=5$, above $\beta$-VAE (0.4786) and $\beta$-VampPrior (0.4848), and its total-correlation versus distortion trade-off is better than both; on skin-lesion and ECG data, the supervised variant cIBP-VAE improves task accuracy over a supervised CNN and over c-VAE, and learns 'triggering' latent units that localize nuisance factors such as lesion color or pacing artifacts.

Load-bearing premise

The results rest on the assumption that, after the Concrete and Kumaraswamy relaxations, the elementwise product of the binary mask and the Gaussian magnitudes still yields latent dimensions that are truly independent, and that a truncated finite number of features behaves like an unbounded process on the datasets used; if either fails, the improved scores could come from extra capacity alone rather than from the independence-preserving growth the paper describes.

Editorial extensions

If this is right

IBP-VAE should keep an advantage over Gaussian-prior VAEs as the number or complexity of generative factors grows, because the latent factor count is inferred rather than preset.
Combining unsupervised nuisance disentangling with a supervised task (cIBP-VAE) should help in domains with many unlabeled nuisance factors, such as medical images or signals, where a plain supervised network struggles.
The learned triggering units provide a concrete interpretability mechanism: deactivating one latent unit removes a single semantic factor (font color, leg style, lesion color, ECG pacing artifact) from reconstructions.
The total-correlation versus distortion analysis is a reusable diagnostic for whether a VAE's posterior capacity is limiting disentanglement.

Reading between the lines

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

A direct testable extension is to sweep the truncation level $K$: if the IBP's infinity is doing the work, performance should improve with $K$ up to a point and then plateau, whereas a finite independent Bernoulli mask of the same total capacity should not match it.
The paper's framing suggests that other nonparametric or flexible independent priors, for example normalizing flows with factorized marginals, could reproduce the same gains, which would show that the growth property rather than the specific IBP is what matters.
In clinical settings, the triggering-unit mechanism could be turned into a hypothesis-generation tool: units that localize lesion color or ECG artifacts can be correlated with patient metadata to find novel confounders, though the paper does not test this.

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

Quick take: this paper is worth a close read for the clinical nuisance-factor experiments and for raising a point people often skip—posterior richness and independence are different axes. But there is a formal problem in the objective. Eq. (6) writes KL(q(Z|ν,X)||p(Z|ν)) with q Concrete and p Bernoulli. Those are not densities on the same space; the KL is infinite as written. The paper never says it relaxes p to a Concrete prior or uses a surrogate KL, such as straight-through with a Bernoulli KL. That means the training objective is not the ELBO they derive, and the "reduced competition" mechanism in Eq. (3) is not what is being optimized. This is load-bearing: the whole story says the IBP prior's independence plus capacity causes the gains, but the actual optimization may be using a different objective or a heuristic gradient estimator.

What is new: prior IBP-VAEs [26,27] focused on reconstruction, not disentanglement. Showing that an IBP prior beats Gaussian and VampPrior on MIG for dSprites, and that a supervised variant separates nuisance factors in skin lesion and ECG data, is a legitimate extension. The TC-D analysis is a nice empirical way to visualize the independence-vs-reconstruction tradeoff. I think the empirical results are plausible; they just are not well supported as written.

Soft spots beyond the KL issue: all quantitative comparisons are single runs without error bars; beta is selected and the best reported; the dSprites comparison to β-TCVAE uses a number from the literature rather than a matched run; no code is released. The clinical experiments are suggestive but small (900 training images, 39 patients). The paper also does not engage Locatello et al. on identifiability, which is a real omission for a claim that richer priors improve disentanglement.

Bottom line: with the ELBO fixed—Concrete prior or explicitly stated Bernoulli KL surrogate—the paper could be a solid contribution. As it stands, the central mechanism is not demonstrated, even though the empirical direction is plausible. I would send it to a serious referee because the clinical material and the independence-vs-capacity question deserve peer time, but I would not cite it in its current form.

Referee Report

4 major / 4 minor

Summary. The paper proposes IBP-VAE, in which the latent representation is generated from a Beta-Bernoulli process via Y = Z ⊙ A, and cIBP-VAE, which augments this model with a supervised classification loss for task-relevant representations. The central hypothesis is that a richer but independent posterior density improves disentangled representation learning by reducing competition between the reconstruction and independence terms in the ELBO. Experiments on colored MNIST, 3D Chairs, dSprites, skin lesion images, and 12-lead ECG report qualitative and quantitative improvements, including a MIG of 0.5477 at β=5 on dSprites versus 0.4786 for β-VAE and 0.4848 for β-VampPrior, and improved clinical classification over supervised baselines.

Significance. If valid, the paper makes a useful conceptual contribution by separating the independence of a latent density from its modeling capacity, and by demonstrating a nonparametric construction that can in principle grow with data complexity. The interpretability analyses on clinical data, including triggering units and label-swap reconstructions, are informative and address an underexplored application area. The paper also provides architectural details and paired statistical tests for the clinical results. However, the theoretical objective as written is not a well-defined ELBO, and the quantitative disentanglement evidence rests on a single benchmark with limited controls, so the central claim is only partially supported.

major comments (4)

[III-C3, Eq. (6)] The KL term KL(q(Z|ν,X) || p(Z|ν)) in Eq. (6) is not finite as written. q(Z|ν,X) is a Concrete distribution on [0,1]^K (Eq. (5) and Appendix A-A), while p(Z|ν) is a product of Bernoulli distributions with point masses at 0 and 1 (Eq. (4)). With respect to any common base measure, the Concrete density is absolutely continuous and the Bernoulli prior is singular, so the KL divergence is infinite or undefined. The paper never states that p(Z|ν) is relaxed to a Concrete prior or that a surrogate KL (e.g., a straight-through estimator with Bernoulli KL) is used. Because the mechanism claimed in Section III-B rests on optimizing the ELBO in Eqs. (3)-(6), this is a load-bearing gap: the actual training objective may differ from the derived one, and the reported improvements cannot be attributed to the IBP ELBO without specifying the actual objective.
[III-C1, IV-A, IV-C] The generative model is defined in the K→∞ limit in Eq. (4), and the paper claims that the latent density 'is able to grow with the complexity of the data' (Section I). However, all experiments use fixed finite truncations: K=100 for colored MNIST, K=50 for the skin lesion data, and K=50 for the ECG data, with no sensitivity analysis over K or the concentration parameter α. A finite truncation with fixed hyperparameters does not by itself establish unbounded growth, and the claim that the model grows with data complexity is not directly tested. Please add truncation-sensitivity experiments or substantially temper the nonparametric-growth claim.
[IV-B, Table I, Fig. 4] The only quantitative disentanglement metric, MIG, is reported on dSprites, with single runs and no standard deviations or significance tests. The values of β are varied over {1,5,10}, but the selection procedure is not described, creating a risk of best-reported selection. In addition, the 'best result reported in literature' for β-TCVAE in Fig. 4 is taken from another paper, so that comparison is not controlled for architecture, latent dimension, or training budget. These limitations do not invalidate the qualitative results or the internal comparison with β-VAE and β-VampPrior, but they weaken the central empirical claim that IBP-VAE improves over state of the art. Please provide error bars, multiple seeds, and ideally an additional benchmark or metric.
[III-C1, III-C2] The claim that the latent representation is independent across dimensions holds for the generative model conditional on ν, but the variational posterior q(Z|ν,X) in Eq. (5) is specified conditional on the shared Kumaraswamy variable ν, and after marginalizing ν and applying the Concrete relaxation the marginal q(Z|X) is not obviously a product of independent factors. If the relaxed binary variables couple dimensions through ν, the 'reduced competition' mechanism in Eq. (3) may not hold for the actual posterior. Please state explicitly whether independence is intended for the conditional posterior q(Z,A|ν,X) or for the marginal q(Z,A|X), and discuss the implications for the stated mechanism.

minor comments (4)

[III-B, Eq. (3)] The index-code mutual information term is written as '-(log N - E_q(y)[H[q(n|y)]]))' with an apparent mismatched parenthesis, and the aggregated posterior q(y) is never explicitly defined as (1/N) Σ_n q(y|x_n). Please define q(y) and correct the formula.
[Fig. 2 caption] The caption lists subfigures (a)-(e) in the text but also refers to panel (f), and the list in the caption is incomplete. Please make the caption consistent with the figure.
[Table II caption] The caption contains a typo: 'classiﬁcation accuray' should be 'classification accuracy'.
[IV-B, Fig. 5] The TC-D analysis does not state how the total-correlation term is estimated from the trained models. Without an explicit estimator, the curves in Fig. 5 are not reproducible; please add the estimation procedure.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity identified: the central claim is an empirical hypothesis tested against external metrics and benchmarks, and no prediction reduces to its own inputs by construction.

full rationale

The paper's central assertion is that an independent latent-factor model with increased posterior capacity improves disentanglement, and this is investigated by comparing IBP-VAE against beta-VAE and beta-VampPrior on external benchmarks. The quantitative disentanglement measure, MIG, is computed from ground-truth factors of variation and is not a term in the optimized objective; the TC-D analysis is a post-hoc decomposition of learned latents, not a training objective. The ELBO decomposition in Eq. (3) is borrowed from prior work and used only as motivation, not as the metric that confirms the claim. Some self-citations exist: [32] is a same-group prior paper used for ECG preprocessing, and [17] is a dataset paper co-authored by two clinical co-authors, but neither is load-bearing for the central derivation. The Concrete-versus-Bernoulli KL in Eq. (6) is a correctness concern about the written objective, not a circular reduction of a prediction to its inputs. No fitted parameter is renamed as a prediction, no uniqueness theorem is imported from the authors' prior work, and no known result is merely relabeled. Accordingly, no significant circularity is present.

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

The model introduces no new physical or conceptual entities beyond a nonparametric Bayesian prior choice and observed 'triggering units' in the latent space. The central claim rests on the factorized independence assumption, the validity of the ELBO decomposition as a mechanism, and tuned hyperparameters, none of which are proved formally.

free parameters (3)

alpha (IBP concentration hyperparameter) = 30 (colored MNIST), 10 (dSprites), 20 (ECG)
Controls the expected number of latent features, which directly sets the capacity the paper argues drives disentanglement. It is chosen per dataset rather than derived from first principles.
beta (KL regularization coefficient) = 1, 5, 10 (grid; best reported at 5)
Weights all KL terms in the ELBO. Selection on validation sets, and on the dSprites training data, affects the reported MIG values.
zeta (supervised loss weight in cIBP-VAE) = 5 (skin, with warm-up); {0.5, 0.8, 1} (ECG, best selected)
Balances the generative and discriminative objectives in the clinical semi-supervised experiments and is tuned per dataset.

assumptions (3)

standard math The ELBO decomposition in Eq. (3) separates total correlation, dimension-wise KL, and index-code mutual information so that these terms can be treated as independently controllable.
The decomposition itself is standard variational calculus, but the paper uses it as a causal argument that a richer density 'lifts competition' between reconstruction and disentanglement; that causal step is a heuristic, not a theorem.
domain assumption Y = Z⊙A is independent across feature dimensions because Z and A are each independent per dimension.
The generative model in Eq. (4) and variational family in Eq. (5) are factorized, so independence is built in through the chosen prior and posterior rather than learned. The disentanglement benefit is attributed to this construction.
domain assumption A truncated stick-breaking construction with Concrete and Kumaraswamy relaxations accurately approximates the infinite IBP and preserves the claimed capacity growth.
Practical runs use K=100 or 50 plus continuous relaxations. The paper does not verify that results are insensitive to truncation level or Concrete temperature.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Improving Disentangled Representation Learning with the Beta Bernoulli Process." pith.science (2026). https://pith.science/paper/DQEABYJQ

@misc{pith2026190901839,
  author       = {Pith},
  title        = {Pith review of: Improving Disentangled Representation Learning with the Beta Bernoulli Process},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/DQEABYJQ}},
  note         = {Machine review of arXiv:1909.01839}
}

read the original abstract

To improve the ability of VAE to disentangle in the latent space, existing works mostly focus on enforcing independence among the learned latent factors. However, the ability of these models to disentangle often decreases as the complexity of the generative factors increases. In this paper, we investigate the little-explored effect of the modeling capacity of a posterior density on the disentangling ability of the VAE. We note that the independence within and the complexity of the latent density are two different properties we constrain when regularizing the posterior density: while the former promotes the disentangling ability of VAE, the latter -- if overly limited -- creates an unnecessary competition with the data reconstruction objective in VAE. Therefore, if we preserve the independence but allow richer modeling capacity in the posterior density, we will lift this competition and thereby allow improved independence and data reconstruction at the same time. We investigate this theoretical intuition with a VAE that utilizes a non-parametric latent factor model, the Indian Buffet Process (IBP), as a latent density that is able to grow with the complexity of the data. Across three widely-used benchmark data sets and two clinical data sets little explored for disentangled learning, we qualitatively and quantitatively demonstrated the improved disentangling performance of IBP-VAE over the state of the art. In the latter two clinical data sets riddled with complex factors of variations, we further demonstrated that unsupervised disentangling of nuisance factors via IBP-VAE -- when combined with a supervised objective -- can not only improve task accuracy in comparison to relevant supervised deep architectures but also facilitate knowledge discovery related to task decision-making. A shorter version of this work will appear in the ICDM 2019 conference proceedings.

Figures

Figures reproduced from arXiv: 1909.01839 by the authors.

**Figure 1.** Outline of the presented IBP-VAE (blue) for unsupervised disentangled representation learning, and cIBP-VAE (blue and orange) for the combination with supervised task learning. 2) Inference model: We introduce a variational approximation of the posterior density qφ(Z, A, ν|X, a, b): qφ(Z, A, ν|X, a, b) = q(A|X)q(Z|ν, X)q(ν|a, b) q(A|X) = Y N n=1 N (µ(xn), diag(σ 2 (xn))) q(Z|ν, X) = Y N n=1 Concrete(π, d(xn)), πk … view at source ↗

**Figure 3.** (Top) Learned latent variables using β-VAE and β-IBP-VAE for the traversal range of (-1, 1). (Bottom) Triggering capacity of the β-IBP-VAE where the three rows give examples of the original images, reconstructed images, and reconstructed images with the triggering unit for leg styles de-activated. Given the diversity of data sets being considered, we leave data and implementation details to each subsection. In all e… view at source ↗

**Figure 4.** [Best viewed in color] Disentangling performance of β-IBP-VAE compared with β-VAE and β-VampPrior at β = 5. The best result reported in literature (rightmost plot, reprinted from β-TCVAE [3] with permission) is also presented for comparison. Each plot shows the relationship between each learned latent dimension (row) and each ground-truth factor (column): color in column one encodes high to low values from blue to r… view at source ↗

Figures from the paper (6 more)

**Figure 5.** Figure 5: TC-D analyses for β-VAE, -IBP-VAE, and -VampPrior on dSprites. The (TC, D) value obtained from each model is plotted for three different β values. in the Appendix B), and we varied the value of β for the penalizing KL terms. A learning rate of 5e-4 and α value of 10 we…

**Figure 6.** Figure 6: ROC curves of cIBP-VAE in comparison to alternative models for classification of melanoma and benign lesions. skin diseases released to support the development of melanoma diagnosis algorithms. Here, we considered the task of classification of dermoscopic images into …

**Figure 7.** Figure 7: [Best viewed in color] (a) Reconstruction examples of cIBP-VAE and cVAE, along with MSE values. (b) Column one: original images; column two: difference in reconstruction after switching the lesion label from malignant to benign (or vice versa); column three: overlay of…

**Figure 8.** Figure 8: ROC curves of cIBP-VAE in comparison to alternative models on the clinical ECG data set. was split into training, validation and test set, where no set shared data from the same patient. The network architecture of IBP-VAE was identical to that used on colored-MNIST. F…

**Figure 9.** Figure 9: 12-lead ECG traces where the pacing artifact is highlighted to the left side of the dotted line. (a) An original signal without the pacing artifact. (b) The reconstructed signal using cIBP-VAE. (c) The reconstructed signal using c-VAE. while c-VAE was not. This shows t…

**Figure 10.** Figure 10: [Best viewed in color] (a): Swapping between the task and nuisance representation from two samples (left) transferred the presence and absence of pacing artifact in the reconstructed signals (right).(b)-(d): Reconstructions before (c) and after (d) de-activation of th…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

43 extracted references · 29 canonical work pages

[1]

Interpretability of deep learning models: a survey of results,

Supriyo Chakraborty, Richard Tomsett, Ramya Raghavendra, Daniel Harborne, Moustafa Alzantot, Federico Cerutti, Mani Srivastava, Alun Preece, Simon Julier, Raghuveer M Rao, et al., “Interpretability of deep learning models: a survey of results,” in IEEE Smart World Congress 2017 Workshop: DAIS, 2017

work page 2017
[2]

beta-vae: Learning basic visual concepts with a constrained variational framework,

Irina Higgins, Loic Matthey, Arka Pal, Christopher Burgess, Xavier Glorot, Matthew Botvinick, Shakir Mohamed, and Alexander Lerchner, “beta-vae: Learning basic visual concepts with a constrained variational framework,” 2016. 10 Reconstruction (c) (d) Turning off the triggering unitOriginal Sample (b) Task Nuisance Task Nuisance (a) Fig. 10: [Best viewed i...

work page 2016
[3]

Isolating sources of disentanglement in variational autoencoders,

Tian Qi Chen, Xuechen Li, Roger B Grosse, and David K Duvenaud, “Isolating sources of disentanglement in variational autoencoders,” in Advances in Neural Information Processing Systems 31 . 2018

work page 2018
[4]

Disentangling by factorising,

Hyunjik Kim and Andriy Mnih, “Disentangling by factorising,” in International Conference on Machine Learning , 2018, pp. 2654–2663

work page 2018
[5]

Vari- ational inference of disentangled latent concepts from unlabeled obser- vations,

Abhishek Kumar, Prasanna Sattigeri, and Avinash Balakrishnan, “Vari- ational inference of disentangled latent concepts from unlabeled obser- vations,” arXiv preprint arXiv:1711.00848 , 2017

arXiv 2017
[6]

Infogan: Interpretable representation learning by information maximizing generative adversarial nets,

Xi Chen, Yan Duan, Rein Houthooft, John Schulman, Ilya Sutskever, and Pieter Abbeel, “Infogan: Interpretable representation learning by information maximizing generative adversarial nets,” in Advances in Neural Information Processing Systems , 2016, pp. 2172–2180

work page 2016
[7]

Auto-encoding variational bayes,

Diederik P Kingma and Max Welling, “Auto-encoding variational bayes,” in Proceedings of the 2nd International Conference on Learning Representations (ICLR), 2013, number 2014

work page 2013
[8]

Stochastic backpropagation and approximate inference in deep gen- erative models,

Danilo Jimenez Rezende, Shakir Mohamed, and Daan Wierstra, “Stochastic backpropagation and approximate inference in deep gen- erative models,” in Proceedings of the 31th International Conference on Machine Learning, ICML 2014, Beijing, China, 21-26 June 2014 , 2014, pp. 1278–1286

work page 2014

Show all 43 references

[9]

V AE with a vampprior,

Jakub M. Tomczak and Max Welling, “V AE with a vampprior,” in International Conference on Artiﬁcial Intelligence and Statistics, AISTATS, 2018

2018
[10]

Improved variational inference with in- verse autoregressive ﬂow,

Durk P Kingma, Tim Salimans, Rafal Jozefowicz, Xi Chen, Ilya Sutskever, and Max Welling, “Improved variational inference with in- verse autoregressive ﬂow,” inAdvances in neural information processing systems, 2016, pp. 4743–4751

2016
[11]

The indian buffet process: An introduction and review,

Thomas L Grifﬁths and Zoubin Ghahramani, “The indian buffet process: An introduction and review,” Journal of Machine Learning Research , vol. 12, no. Apr, pp. 1185–1224, 2011

2011
[12]

Gradient-based learning applied to document recognition,

Yann LeCun, L ´eon Bottou, Yoshua Bengio, and Patrick Haffner, “Gradient-based learning applied to document recognition,” Proceedings of the IEEE , vol. 86, no. 11, pp. 2278–2324, 1998

1998
[13]

Seeing 3d chairs: exemplar part-based 2d-3d alignment using a large dataset of cad models,

Mathieu Aubry, Daniel Maturana, Alexei A Efros, Bryan C Russell, and Josef Sivic, “Seeing 3d chairs: exemplar part-based 2d-3d alignment using a large dataset of cad models,” in Proceedings of the IEEE conference on computer vision and pattern recognition, 2014, pp. 3762– 3769

2014
[14]

dsprites: Disentanglement testing sprites dataset,

Loic Matthey, Irina Higgins, Demis Hassabis, and Alexander Lerchner, “dsprites: Disentanglement testing sprites dataset,” https://github.com/deepmind/dsprites-dataset/, 2017

2017
[15]

David Gutman, Noel CF Codella, Emre Celebi, Brian Helba, Michael Marchetti, Nabin Mishra, and Allan Halpern, “Skin lesion analysis toward melanoma detection: A challenge at the international symposium on biomedical imaging (isbi) 2016, hosted by the international skin imaging ...

2016 arXiv
[16]

Giuseppe Argenziano, Gabriella Fabbrocini, Paolo Carli, Vincenzo De Giorgi, Elena Sammarco, and Mario Delﬁno, “Epiluminescence microscopy for the diagnosis of doubtful melanocytic skin lesions: comparison of the abcd rule of dermatoscopy and a new 7-point checklist based on pa...

1998
[17]

Real-time localization of ventricular tachycardia origin from the 12-lead electrocardiogram,

John L Sapp, Meir Bar-Tal, Adam J Howes, Jonathan E Toma, Ahmed El-Damaty, James W Warren, Paul J MacInnis, Shijie Zhou, and B Milan Hor´aˇcek, “Real-time localization of ventricular tachycardia origin from the 12-lead electrocardiogram,” JACC: Clinical Electrophysiology, vol....

2017
[18]

Gen- erative adversarial nets,

Ian Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair, Aaron Courville, and Yoshua Bengio, “Gen- erative adversarial nets,” in Advances in neural information processing systems, 2014, pp. 2672–2680

2014
[19]

Elbo surgery: yet another way to carve up the variational evidence lower bound,

Matthew D Hoffman and Matthew J Johnson, “Elbo surgery: yet another way to carve up the variational evidence lower bound,” in Workshop in Advances in Approximate Bayesian Inference, NIPS , 2016

2016
[20]

Fixing a broken elbo,

Alexander Alemi, Ben Poole, Ian Fischer, Joshua Dillon, Rif A Saurous, and Kevin Murphy, “Fixing a broken elbo,” in International Conference on Machine Learning , 2018, pp. 159–168

2018
[21]

Importance weighted autoencoders,

Yuri Burda, Roger Grosse, and Ruslan Salakhutdinov, “Importance weighted autoencoders,” arXiv preprint arXiv:1509.00519 , 2015

2015 arXiv
[22]

Deep unsupervised clustering with gaussian mixture variational autoencoders,

Nat Dilokthanakul, Pedro AM Mediano, Marta Garnelo, Matthew CH Lee, Hugh Salimbeni, Kai Arulkumaran, and Murray Shanahan, “Deep unsupervised clustering with gaussian mixture variational autoencoders,” arXiv preprint arXiv:1611.02648 , 2016

2016 arXiv
[23]

Variational lossy autoencoder,

Xi Chen, Diederik P Kingma, Tim Salimans, Yan Duan, Prafulla Dhari- wal, John Schulman, Ilya Sutskever, and Pieter Abbeel, “Variational lossy autoencoder,” arXiv preprint arXiv:1611.02731 , 2016

2016 arXiv
[24]

Stick-breaking variational autoen- coders,

Eric Nalisnick and Padhraic Smyth, “Stick-breaking variational autoen- coders,” arXiv preprint arXiv:1605.06197 , 2016

2016 arXiv
[25]

Nonparametric variational auto-encoders for hierarchical representation learning.,

Prasoon Goyal, Zhiting Hu, Xiaodan Liang, Chenyu Wang, Eric P Xing, and Carnegie Mellon, “Nonparametric variational auto-encoders for hierarchical representation learning.,” in ICCV, 2017, pp. 5104–5112

2017
[26]

Indian buffet process deep generative models,

Sotirios P Chatzis, “Indian buffet process deep generative models,” arXiv preprint arXiv:1402.3427, 2014

2014 arXiv
[27]

Structured varia- tional autoencoders for the beta-bernoulli process,

Rachit Singh, Jeffrey Ling, and Finale Doshi-Velez, “Structured varia- tional autoencoders for the beta-bernoulli process,”NIPS 2017 Workshop on Advances in Approximate Bayesian Inference , 2017

2017
[28]

Learning fair representations,

Rich Zemel, Yu Wu, Kevin Swersky, Toni Pitassi, and Cynthia Dwork, “Learning fair representations,” in International Conference on Machine Learning, 2013

2013
[29]

The variational fair autoencoder,

Christos Louizos, Kevin Swersky, Yujia Li, Max Welling, and Richard Zemel, “The variational fair autoencoder,” arXiv preprint arXiv:1511.00830, 2015

2015 arXiv
[30]

Fair deep learning prediction for healthcare applications with confounder ﬁltering,

Zhenglin Wu, Haohan Wang, Mingze Cao, Yin Chen, and Eric P Xing, “Fair deep learning prediction for healthcare applications with confounder ﬁltering,” arXiv preprint arXiv:1803.07276 , 2018

2018 arXiv
[31]

Stick-breaking construction for the indian buffet process,

Yee Whye Teh, Dilan Gr ¨ur, and Zoubin Ghahramani, “Stick-breaking construction for the indian buffet process,” in Artiﬁcial Intelligence and Statistics, 2007, pp. 556–563

2007
[32]

Disentangling inter-subject variations: Automatic localization of ventricular tachycardia origin from 12-lead electrocardiograms,

Shuhang Chen, Prashnna K Gyawali, Huafeng Liu, B Milan Horacek, John L Sapp, and Linwei Wang, “Disentangling inter-subject variations: Automatic localization of ventricular tachycardia origin from 12-lead electrocardiograms,” in Biomedical Imaging (ISBI 2017), 2017 IEEE 14th I...

2017
[33]

Categorical reparameterization with gumbel-softmax,

Eric Jang, Shixiang Gu, and Ben Poole, “Categorical reparameterization with gumbel-softmax,” arXiv preprint arXiv:1611.01144 , 2016

2016 arXiv
[34]

The concrete distribution: A continuous relaxation of discrete random variables,

Chris J Maddison, Andriy Mnih, and Yee Whye Teh, “The concrete distribution: A continuous relaxation of discrete random variables,” arXiv preprint arXiv:1611.00712 , 2016

2016 arXiv
[35]

Semi-supervised learning with deep generative models,

Diederik P Kingma, Shakir Mohamed, Danilo Jimenez Rezende, and Max Welling, “Semi-supervised learning with deep generative models,” 11 in Advances in Neural Information Processing Systems, 2014, pp. 3581– 3589

2014
[36]

Adam: A method for stochastic optimization,

Diederik Kingma and Jimmy Ba, “Adam: A method for stochastic optimization,” arXiv preprint arXiv:1412.6980 , 2014

2014 arXiv
[37]

Three-point checklist of dermoscopy,

H Peter Soyer, Giuseppe Argenziano, Iris Zalaudek, Rosamaria Corona, Francesco Sera, Renato Talamini, Filomena Barbato, Adone Baroni, Lorenza Cicale, Alessandro Di Stefani, et al., “Three-point checklist of dermoscopy,” Dermatology, vol. 208, no. 1, pp. 27–31, 2004

2004
[38]

Skin lesion analysis towards melanoma detection using deep learning network,

Yuexiang Li and Linlin Shen, “Skin lesion analysis towards melanoma detection using deep learning network,” Sensors, vol. 18, no. 2, pp. 556, 2018

2018
[39]

One weird trick for parallelizing convolutional neural networks,

Alex Krizhevsky, “One weird trick for parallelizing convolutional neural networks,” arXiv preprint arXiv:1404.5997 , 2014

2014 arXiv
[40]

Generating sentences from a continuous space,

Samuel R Bowman, Luke Vilnis, Oriol Vinyals, Andrew M Dai, Rafal Jozefowicz, and Samy Bengio, “Generating sentences from a continuous space,” Proceedings of the 20th SIGNLL Conference on Computational Natural Language Learning, CoNLL , 2016

2016
[41]

Robert Plonsey, Bioelectric phenomena, Wiley Online Library, 1969

1969
[42]

Localization of origins of premature ventricular contraction by means of convolutional neural network from 12-lead ecg,

Ting Yang, Long Yu, Qi Jin, Liqun Wu, and Bin He, “Localization of origins of premature ventricular contraction by means of convolutional neural network from 12-lead ecg,” IEEE Transactions on Biomedical Engineering, vol. 65, no. 7, pp. 1662–1671, 2018

2018
[43]

Challenging common assumptions in the unsupervised learning of disentangled representations,

Francesco Locatello, Stefan Bauer, Mario Lucic, Sylvain Gelly, Bernhard Sch¨olkopf, and Olivier Bachem, “Challenging common assumptions in the unsupervised learning of disentangled representations,” arXiv preprint arXiv:1811.12359, 2018

2018 arXiv

Pith tools

Reviewed August 14, 2026 · model on record in the stance chip above.

[1] [1]

Interpretability of deep learning models: a survey of results,

Supriyo Chakraborty, Richard Tomsett, Ramya Raghavendra, Daniel Harborne, Moustafa Alzantot, Federico Cerutti, Mani Srivastava, Alun Preece, Simon Julier, Raghuveer M Rao, et al., “Interpretability of deep learning models: a survey of results,” in IEEE Smart World Congress 2017 Workshop: DAIS, 2017

work page 2017

[2] [2]

beta-vae: Learning basic visual concepts with a constrained variational framework,

Irina Higgins, Loic Matthey, Arka Pal, Christopher Burgess, Xavier Glorot, Matthew Botvinick, Shakir Mohamed, and Alexander Lerchner, “beta-vae: Learning basic visual concepts with a constrained variational framework,” 2016. 10 Reconstruction (c) (d) Turning off the triggering unitOriginal Sample (b) Task Nuisance Task Nuisance (a) Fig. 10: [Best viewed i...

work page 2016

[3] [3]

Isolating sources of disentanglement in variational autoencoders,

Tian Qi Chen, Xuechen Li, Roger B Grosse, and David K Duvenaud, “Isolating sources of disentanglement in variational autoencoders,” in Advances in Neural Information Processing Systems 31 . 2018

work page 2018

[4] [4]

Disentangling by factorising,

Hyunjik Kim and Andriy Mnih, “Disentangling by factorising,” in International Conference on Machine Learning , 2018, pp. 2654–2663

work page 2018

[5] [5]

Vari- ational inference of disentangled latent concepts from unlabeled obser- vations,

Abhishek Kumar, Prasanna Sattigeri, and Avinash Balakrishnan, “Vari- ational inference of disentangled latent concepts from unlabeled obser- vations,” arXiv preprint arXiv:1711.00848 , 2017

arXiv 2017

[6] [6]

Infogan: Interpretable representation learning by information maximizing generative adversarial nets,

Xi Chen, Yan Duan, Rein Houthooft, John Schulman, Ilya Sutskever, and Pieter Abbeel, “Infogan: Interpretable representation learning by information maximizing generative adversarial nets,” in Advances in Neural Information Processing Systems , 2016, pp. 2172–2180

work page 2016

[7] [7]

Auto-encoding variational bayes,

Diederik P Kingma and Max Welling, “Auto-encoding variational bayes,” in Proceedings of the 2nd International Conference on Learning Representations (ICLR), 2013, number 2014

work page 2013

[8] [8]

Stochastic backpropagation and approximate inference in deep gen- erative models,

Danilo Jimenez Rezende, Shakir Mohamed, and Daan Wierstra, “Stochastic backpropagation and approximate inference in deep gen- erative models,” in Proceedings of the 31th International Conference on Machine Learning, ICML 2014, Beijing, China, 21-26 June 2014 , 2014, pp. 1278–1286

work page 2014