Pith. sign in

REVIEW 3 major objections 4 minor 54 references

Deep Generative Models with Hard Linear Equality Constraints

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

Pith's one-line read Conditioning a generative model's latent distribution on a linear equality constraint satisfies it exactly and keeps or improves data fidelity.

desk verdict A broad empirical study of a simple idea—conditioning the Gaussian on the constraint—with a theoretically under-supported gradient estimator that needs reframing or a bias bound. read the letter →

arxiv 2502.05416 v2 pith:XGGM3U3M submitted 2025-02-08 cs.LG

classification cs.LG
keywords constrainedgenerativemodelinghardlinearequalityconstraintsGaussianconditionaldistributionsgradientestimatorsneuro-symboliclearningvariationalautoencodersdiffusionmodelsscientificmachine
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 claims that deep generative models can be made to respect hard linear equality constraints—charge neutrality in molecules, mass balance in chemical plants, budget allocation in portfolios—by conditioning the model's distribution on the constraint rather than by sampling freely and then patching the samples. It documents that ordinary generative models almost surely violate such constraints even when trained on compliant data, and that existing projection-based fixes restore compliance at the cost of data fidelity. The proposed approach instead trains the constrained distribution $z \sim p_\theta(z \mid Az = k)$ end-to-end, using a gradient estimator that differentiates through the conditional marginal expectation as a differentiable proxy for the latent. Across VAEs, diffusion models, and graph neural networks on five image datasets and three scientific applications, the method attains zero constraint violations while matching or beating unconstrained models and every baseline on fidelity and accuracy metrics. If the paper is right, any Gaussian-latent generative model can absorb linear domain knowledge exactly, with no post-hoc correction layer and no fidelity penalty.

What carries the argument

The central object is the constrained Gaussian $z \sim p_\theta(z \mid Az = k)$: the distribution of a Gaussian latent conditioned on a linear equality, which is itself Gaussian with closed-form parameters, mean $\mu + \Sigma A^\top (A\Sigma A^\top)^{-1}(k - A\mu)$ and covariance $\Sigma - \Sigma A^\top (A\Sigma A^\top)^{-1} A \Sigma$. This identity converts constraint satisfaction from a post-hoc projection step into an exact sampling step, usable anywhere in a generative model. The load-bearing training mechanism is the gradient approximation $\nabla_\theta L(x,y;\omega) \approx \partial_\theta m(\theta)\, \nabla_z \ell(x,y;\omega)$, with $m(\theta)$ chosen as the conditional marginal expectation $\mathbb{E}[z_i \mid Az = k]$; this proxy is differentiable, carries constraint information, and is shown empirically to dominate constrained marginals, a constrained reparameterization trick, and Constrained Layer projections on bias, variance, and error. A second mechanism, the closed-form expected loss for identity decoders with $\ell^1$ or $\ell^2$ losses, removes the need for gradient estimation in regression-style settings and powers the paper's strongest results on scientific applications.

What would settle it

Run the brightness-constrained MNIST VAE with the paper's marginal-expectation proxy replaced by a score-function (REINFORCE) estimator of the exact constrained gradient, holding everything else fixed; if the two training runs converge to materially different latents or the score-function run achieves higher held-out likelihood, the proxy is biased and the reported gains are estimator-specific rather than inherent to distribution-level constraint enforcement.

Watch

Extended reading notes

Core claim

The paper's central claim is that a hard linear equality constraint $Az = k$ should be enforced at the level of the distribution, not the sample. Given an encoder that outputs Gaussian parameters $\theta = (\mu, \Sigma)$, the model conditions the latent directly: $z \sim p_\theta(z \mid Az = k)$, a Gaussian whose mean and covariance have closed forms, so exact sampling from the feasible set costs no extra machinery. Training targets the constrained expected loss $\mathbb{E}[\ell(f_u(z), y) \mid Az = k]$, and since the gradient of that expectation has no closed form in general, the paper backpropagates through the proxy $\nabla_\theta L \approx \partial_\theta m(\theta)\, \nabla_z \ell(f_u(z), \hat{y})$ with $m(\theta) = \mathbb{E}[z_i \mid Az = k]$, the conditional marginal expectation. A controlled comparison on a synthetic Gaussian task shows this proxy to have far lower bias and average error than alternatives, including conditional marginal densities, a variance-weighted constrained reparameterization, and projection-based Constrained Layers. When the decoder is the identity map and the loss is $\ell^1$ or $\ell^2$, the constrained expected loss is available in closed form, eliminating the estimator altogether. The paper then shows, on brightness-constrained MNIST, CIFAR-10, CelebA, and LSUN images and on MOF charge prediction, chemical-process surrogates, and stock allocation, that unconstrained models violate the constraints with probability close to one while the constrained-distribution method achieves zero violations and better or equal fidelity scores in every reported setting.

Load-bearing premise

The load-bearing premise is that the gradient of the constrained expected loss is faithfully approximated by differentiating the conditional marginal expectation (the paper's Equation 5); this proxy is validated empirically on a synthetic Gaussian task with an identity decoder and $\ell^1$ or $\ell^2$ losses, but it is not derived or proven unbiased for nonlinear decoders, general loss functions, or large constraint matrices.

Editorial extensions

If this is right

  • Any deep generative model with a Gaussian latent or output layer can satisfy a linear equality exactly at train and test time, replacing the projection step with a closed-form mean-and-covariance update that adds under a second per epoch in the VAE experiments.
  • Diffusion models can insert this exact constrained sampling at selected backward steps; on CIFAR-10, CelebA, LSUN Church, and LSUN Cat, the constrained DDPM and DDIM variants drive the violation rate from roughly 0.999 to 0 while matching or improving FID and Inception Score.
  • In identity-decoder regression settings, the closed-form Gaussian expected loss trains without Monte Carlo noise and beats projection baselines on MOF charge prediction, chemical-process surrogates (CSTR scaled MSE 4.31 versus 9.28), and Sharpe-ratio portfolio allocation (1.9041 versus 1.8092).
  • Unconstrained generative models do not absorb linear constraints from compliant training data; the paper measures violation rates near 1.0 for plain VAEs and diffusion models, so explicit distribution-level enforcement is necessary rather than optional.
  • Across estimator designs, the conditional marginal expectation beats the conditional marginal density as a continuous-domain proxy, a divergence the paper links to the discrete k-subset setting where the two coincide.

Reading between the lines

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

  • A stress test the paper does not run: how the marginal-expectation proxy behaves with large or block-structured constraint matrices, since the synthetic validation used small Gaussian instances and the estimator's bias scaling is unexamined.
  • The conditioning identity suggests a general recipe — any latent family with tractable $p(z \mid Az = k)$ and tractable conditional marginals could inherit the same end-to-end training; the paper's Poisson and multinomial results are a first step toward that generalization.
  • An implicit consequence the paper leaves unanalyzed: conditioning on $a$ linear equalities restricts the latent to an $(n-a)$-dimensional affine subspace, so very tight constraint systems shrink effective latent dimensionality, which could matter when the constraint count approaches the latent dimension.
  • Since the proxy-fidelity evidence uses an identity decoder, a decisive follow-up would compare the proxy against a score-function estimator on a nonlinear decoder; divergent gradient directions would re-attribute the reported gains to the constrained-distribution formulation rather than to the estimator itself.
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 paper proposes to enforce hard linear equality constraints in deep generative models by constraining the Gaussian latent or output distribution itself, i.e., sampling from pθ(z | Az = k) rather than drawing from an unconstrained distribution and projecting post hoc. It derives exact sampling from the Gaussian conditional and closed-form conditional marginals, proposes several gradient estimators with "marginal expectation" as the preferred proxy, and derives closed-form expected L1/L2 losses for the identity-decoder case. The method is evaluated across VAEs, diffusion models, an MPNN for MOF charge prediction, chemical process surrogates, and stock allocation, with the central claim that the proposed approach guarantees constraint satisfaction and achieves superior generative performance on every benchmark.

Significance. If the central claims held, this would be a practically valuable and architecture-agnostic alternative to post-hoc projection baselines: exact constrained Gaussian sampling and closed-form conditionals are clean and useful ingredients, and the breadth of experiments is a real asset. The paper also provides a Poisson extension in Appendix G, which broadens its scope. However, the theoretical support for the main gradient estimator is incomplete, and the closed-form loss formulas contain algebra/notation problems that affect the synthetic ground truth and several experiments. The significance is therefore contingent on fixing these load-bearing issues.

major comments (3)
  1. [§4.1, Eq. (5)] The proxy ∇θL ≈ ∂θm(θ)∇zℓ is stated without derivation or an error bound, and for the exact sampling procedure it discards all dependence on the covariance of the constrained Gaussian. Writing the exact sample as z = μc + Lcε, the true gradient contains ∂θLc-type terms; for a nonlinear decoder these terms are first-order in θ and are not negligible. The synthetic validation in §4.2 covers only identity decoders with L1/L2 losses, so it cannot justify the use of this estimator with the sigmoid VAE, U-Net, or MPNN decoders used in §6, nor with NLL/ELBO objectives. Please provide either a bias analysis/proof for Eq. (5) or an empirical validation on at least one nonlinear decoder using the actual objectives of the paper.
  2. [§5, Prop. 5.1; Appendix H.2] Under the paper's definition of Σ as a covariance matrix, the closed-form L1 and L2 expected losses are not correct as printed. For L2, E[(zi − yi)^2] = μ_i^2 + Σ_{i,i} − 2y_i μ_i + y_i^2, not μ_i^2 + Σ_{i,i}^2 − 2y_i μ_i + y_i^2; for L1, the coefficient of the exponential term must be the standard deviation, i.e., σ√(2/π) with σ = √Σ_{i,i}, not the variance. The proof in H.2 displays a term of the form μ_i + σ_i^4 − 2y_i μ_i + y_i^2, which is not the second moment of a Gaussian. Because these closed forms are used both as the ground truth for the estimator comparison in Figure 3 and as training losses in §6.4 and §6.5, this is load-bearing; the formulas must be corrected or the notation must be redefined consistently and propagated through the proofs and experiments.
  3. [Table 5, Appendix I] The closed-form loss produces NLL > 1e7 in the MOF experiment, and Appendix I shows that this is caused by variance collapse. This is not a harmless numerical artifact: it means the closed-form trainer is effectively optimizing a degenerate distribution whose likelihood is zero, while the explanation notes that MAD is unaffected. The paper should qualify the claim of probabilistic soundness, restrict it to the likelihood objective, or add a remedy such as a variance floor and re-evaluate the affected experiments.
minor comments (4)
  1. [Abstract and §6.2] "archives" should be "achieves", and "Reults" should be "Results".
  2. [Table 3] On CIFAR and LSUN Church, the IS of Ours is lower than that of DDPM (9.223 vs 9.278 and 2.471 vs 2.460, respectively), so the abstract's claim of superiority "across every benchmark" should be qualified to the primary metric or to FID.
  3. [Appendix G, Prop. G.2] The expression "Pn j=1 zn" should read "Pn j=1 zj".
  4. [Appendix A] The description of the ground-truth gradients is confusing: "The ground truth are sampled from N(0,I) satisfying the constraint" should be clarified to explain how h_gt is computed for Figure 3.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the constrained sampling and closed-form losses are derived from standard Gaussian identities, and the gradient estimator is an explicitly stated approximation validated against a derived ground truth.

full rationale

The paper's derivation chain is self-contained rather than circular. Equation (1) defines the model as z ~ pθ(z | Az=k); Equation (2) is the corresponding expected loss. Exact sampling from the constrained Gaussian is established in Proposition H.1 via standard conditional-normal formulas. Proposition 4.1 derives conditional marginals and expectations, and Proposition 5.1 derives the L1/L2 closed-form expected losses from those marginals; no parameter is fitted to make these identities match a target result. The gradient estimator in Equation (5) is presented explicitly as an approximation with candidate proxies m(θ), and Section 4.2 compares estimators against ground-truth gradients obtained from the derived closed-form loss, which is an internal but non-circular validation because the closed form is proved rather than assumed. The only overlapping-author citation, Ahmed et al. (2023), is used for the Bernoulli-case equivalence of two estimators and is not load-bearing for the Gaussian derivation, which is carried out independently. Empirical claims compare against external baselines on externally specified datasets; no fitted parameter is renamed as a prediction. Therefore no circularity is found.

Assumptions & free parameters 2 free parameters · 6 assumptions · 0 invented entities

The method rests on standard Gaussian conditioning plus a set of explicit modeling choices. The central new piece, the gradient proxy, is assumed without proof, which is why it is listed as an ad-hoc axiom. No new physical entities or unobserved quantities are introduced.

free parameters (2)
  • Constrained diffusion step count and placement schedule = N=3, Start 3 End 3
    Selected on CIFAR-10 by sweeping N and schedule (Tables 9 and 10) to balance FID and IS, then applied to CelebA, LSUN Church, and LSUN Cat without re-tuning.
  • Constraint constants for synthetic image benchmarks = MNIST: sum of pixels = 100; CIFAR/CelebA/LSUN: per-channel sum = dataset mean or median
    These are hand-chosen constants that define the synthetic benchmarks; they are not fitted to model performance, but they determine the difficulty of the constraint task.
assumptions (6)
  • standard math Gaussian conditioning formula: p(z | Az=k) is Gaussian with mean mu + Sigma A^T (A Sigma A^T)^-1 (k - A mu) and covariance Sigma - Sigma A^T (A Sigma A^T)^-1 A Sigma.
    Used in Proposition H.1 and Proposition 4.1; this is a textbook result for conditioning a multivariate Gaussian on a linear equality.
  • domain assumption The latent or output variable z follows a multivariate Gaussian distribution with parameters produced by a neural encoder.
    Equation 1 in Section 3 assumes z ~ N(mu, Sigma); this is the central modeling assumption that makes exact constrained sampling tractable.
  • domain assumption The constraint matrix A has full row rank a <= n.
    Section 3 states rank(A) = a <= n, which guarantees A Sigma A^T is invertible for positive definite Sigma.
  • domain assumption For the closed-form expected loss, the decoder fu is the identity function and the loss is L1 or L2.
    Section 5 introduces this assumption to derive closed forms; it limits the scope of that part of the theory.
  • ad hoc to paper The gradient of the constrained expected loss can be approximated as partial theta m(theta) times gradient of the loss with respect to z, where m is the conditional marginal density or expectation.
    Section 4.1, Equation 5, states this approximation without derivation or convergence guarantee; it is the main heuristic the empirical claims rest on.
  • standard math For Poisson variables, conditioning on the sum k yields a multinomial distribution and binomial marginals.
    Used in Appendix G for the discrete extension; standard Poisson conditioning results.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Deep Generative Models with Hard Linear Equality Constraints." pith.science (2026). https://pith.science/paper/XGGM3U3M

@misc{pith2026250205416,
  author       = {Pith},
  title        = {Pith review of: Deep Generative Models with Hard Linear Equality Constraints},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/XGGM3U3M}},
  note         = {Machine review of arXiv:2502.05416}
}
read the original abstract

While deep generative models~(DGMs) have demonstrated remarkable success in capturing complex data distributions, they consistently fail to learn constraints that encode domain knowledge and thus require constraint integration. Existing solutions to this challenge have primarily relied on heuristic methods and often ignore the underlying data distribution, harming the generative performance. In this work, we propose a probabilistically sound approach for enforcing the hard constraints into DGMs to generate constraint-compliant and realistic data. This is achieved by our proposed gradient estimators that allow the constrained distribution, the data distribution conditioned on constraints, to be differentiably learned. We carry out extensive experiments with various DGM model architectures over five image datasets and three scientific applications in which domain knowledge is governed by linear equality constraints. We validate that the standard DGMs almost surely generate data violating the constraints. Among all the constraint integration strategies, ours not only guarantees the satisfaction of constraints in generation but also archives superior generative performance than the other methods across every benchmark.

Figures

Figures reproduced from arXiv: 2502.05416 by the authors.

Figure 1
Figure 1. Comparison of different methods for generating samples that satisfy linear equality constraints. The left panel shows the original unconstrained distribution in a 2-dimensional plane, with the purple line representing the constraint x1 + x2 = 0. Our proposed method generates the most realistic sample as indicated by the right figure, outperforming existing methods that optimize for L1 distance (CL) and L2 distance (… view at source ↗
Figure 2
Figure 2. The constrained model considered in this work. It involves an encoder hv that outputs θ to parameterize a latent distribution constrained by the linear equality constraint Az = k. We first study when the objective admits a closed-form expression such that standard training is amenable. We further propose and study various gradient estimators for the general case by combining exact sampling in the forward pass and gr… view at source ↗
Figure 3
Figure 3. Comparisons of gradient estimators for point-wise loss ℓ being L1 loss (upper plot) and L2 loss (lower plot) applied to Gaussian variable are conducted. To compare the directions of the estimated and ground-truth gradients, we utilize the cosine distance. The bias, variance, and error of the gradient estimators are measured using a sample size of 10, 000. 4.2. Comparison of Gradient Estimators We present a rigorous … view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: Comparison of gradient estimators for VAE with con￾strained latent space. Negative log-likelihood (NLL), negative ELBO (NELBO), and reconstruction loss (RL) are averaged over 5 trials. KL divergence between a constrained approximate poste￾rior pθ (z | Az = k, x) and a …
Figure 5
Figure 5. Figure 5: The first block displays the original MNIST images and the ones modified by the brightness constraint as inputs. For the following blocks, each displays the reconstructed images by differ￾ent VAE architectures. Within each block, the first row is generated by the uncon…
Figure 6
Figure 6. Figure 6: Training and validation MSE loss curve for CSTR. All results are averaged over 10 independent runs [PITH_FULL_IMAGE:figures/full_fig_p015_6.png]
Figure 7
Figure 7. Figure 7: Training and validation MSE loss curve for plant. All results are averaged over 10 independent runs. • y3: Condenser heat duty of COLUMN column. • y4: Reboiler heat duty of COLUMN column. • y5: Condenser heat duty of COL-REC column. • y6: Reboiler heat duty of COL-REC …
Figure 8
Figure 8. Figure 8: Training and validation MSE loss curve for distillation. All results are averaged over 10 independent runs. G. Beyond Gaussian In this section, we present the theoretical results when z are Poisson variables defined over discrete domains. Similar to the Gaussian settin…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

54 extracted references · 42 canonical work pages

  1. [1]

    write newline

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

  2. [2]

    Semantic probabilistic layers for neuro-symbolic learning

    Ahmed, K., Teso, S., Chang, K.-W., Van den Broeck, G., and Vergari, A. Semantic probabilistic layers for neuro-symbolic learning. Advances in Neural Information Processing Systems, 35: 0 29944--29959, 2022

  3. [3]

    Simple: A gradient estimator for k-subset sampling

    Ahmed, K., Zeng, Z., Niepert, M., and Van den Broeck, G. Simple: A gradient estimator for k-subset sampling. In Proceedings of the International Conference on Learning Representations (ICLR), may 2023

  4. [4]

    and Kolter, J

    Amos, B. and Kolter, J. Z. O pt N et: Differentiable optimization as a layer in neural networks. In Proceedings of the 34th International Conference on Machine Learning, volume 70 of Proceedings of Machine Learning Research, pp.\ 136--145. PMLR, 2017

  5. [5]

    d., Serafini, L., and Spranger, M

    Badreddine, S., Garcez, A. d., Serafini, L., and Spranger, M. Logic tensor networks. Artificial Intelligence, 303: 0 103649, 2022

  6. [6]

    Importance weighted autoencoders

    Burda, Y., Grosse, R., and Salakhutdinov, R. Importance weighted autoencoders. In International Conference on Learning Representations (ICLR), 2016

  7. [7]

    and Kwon, R

    Butler, A. and Kwon, R. Integrating prediction in mean-variance portfolio optimization. Available at SSRN 3788875, 2021

  8. [8]

    Spectral temporal graph neural network for multivariate time-series forecasting

    Cao, D., Wang, Y., Duan, J., Zhang, C., Zhu, X., Huang, C., Tong, Y., Xu, B., Bai, J., Tong, J., et al. Spectral temporal graph neural network for multivariate time-series forecasting. Advances in Neural Information Processing Systems, 33: 0 17766--17778, 2020

Show all 54 references
  1. [9]

    Chen, H., Flores, G. E. C., and Li, C. Physics-informed neural networks with hard linear equality constraints. Computers Chemical Engineering, 189: 0 108764, 2024. ISSN 0098-1354. doi:https://doi.org/10.1016/j.compchemeng.2024.108764

  2. [10]

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

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

  3. [11]

    Problog: A probabilistic prolog and its application in link discovery

    De Raedt, L., Kimmig, A., and Toivonen, H. Problog: A probabilistic prolog and its application in link discovery. In IJCAI 2007, Proceedings of the 20th international joint conference on artificial intelligence, pp.\ 2462--2467. IJCAI-INT JOINT CONF ARTIF INTELL, 2007

  4. [12]

    Efficient generation of structured objects with constrained adversarial networks

    Di Liello, L., Ardino, P., Gobbi, J., Morettin, P., Teso, S., and Passerini, A. Efficient generation of structured objects with constrained adversarial networks. Advances in neural information processing systems, 33: 0 14663--14674, 2020

  5. [13]

    and Zabell, S

    Diaconis, P. and Zabell, S. Closed form summation for classical distributions: variations on a theme of de moivre. Statistical Science, pp.\ 284--302, 1991

  6. [14]

    Bridging logic and kernel machines

    Diligenti, M., Gori, M., Maggini, M., and Rigutini, L. Bridging logic and kernel machines. Machine learning, 86: 0 57--88, 2012

  7. [15]

    and Krause, A

    Djolonga, J. and Krause, A. Differentiable learning of submodular models. In Guyon, I., Luxburg, U. V., Bengio, S., Wallach, H., Fergus, R., Vishwanathan, S., and Garnett, R. (eds.), Advances in Neural Information Processing Systems, volume 30. Curran Associates, Inc., 2017

  8. [16]

    Donti, P., Amos, B., and Kolter, J. Z. Task-based end-to-end model learning in stochastic optimization. In Advances in Neural Information Processing Systems, pp.\ 5484--5494, 2017

  9. [17]

    Dl2: training and querying neural networks with logic

    Fischer, M., Balunovic, M., Drachsler-Cohen, D., Gehr, T., Zhang, C., and Vechev, M. Dl2: training and querying neural networks with logic. In International Conference on Machine Learning, pp.\ 1931--1941. PMLR, 2019

  10. [18]

    Garcez, A. d. and Lamb, L. C. Neurosymbolic ai: The 3 rd wave. Artificial Intelligence Review, 56 0 (11): 0 12387--12406, 2023

  11. [19]

    and Lukasiewicz, T

    Giunchiglia, E. and Lukasiewicz, T. Multi-label classification neural networks with hard logical constraints. Journal of Artificial Intelligence Research, 72: 0 759--818, 2021

  12. [20]

    Stochastic optimization of sorting networks via continuous relaxations

    Grover, A., Wang, E., Zweig, A., and Ermon, S. Stochastic optimization of sorting networks via continuous relaxations. In International Conference on Learning Representations, 2019

  13. [21]

    Variational autoencoders with jointly optimized latent dependency structure

    He, J., Gong, Y., Marino, J., Mori, G., and Lehrmann, A. Variational autoencoders with jointly optimized latent dependency structure. In International conference on learning representations, 2018

  14. [22]

    N., Jidling, C., Wills, A

    Hendriks, J. N., Jidling, C., Wills, A. G., and Sch \"o n, T. B. Linearly constrained neural networks. ArXiv, abs/2002.01600, 2020

  15. [23]

    Gans trained by a two time-scale update rule converge to a local nash equilibrium

    Heusel, M., Ramsauer, H., Unterthiner, T., Nessler, B., and Hochreiter, S. Gans trained by a two time-scale update rule converge to a local nash equilibrium. In Advances in Neural Information Processing Systems (NeurIPS), 2017

  16. [24]

    Denoising diffusion probabilistic models

    Ho, J., Jain, A., and Abbeel, P. Denoising diffusion probabilistic models. arXiv preprint arxiv:2006.11239, 2020

  17. [25]

    Categorical reparameterization with gumbel-softmax

    Jang, E., Gu, S., and Poole, B. Categorical reparameterization with gumbel-softmax. In International Conference on Learning Representations, 2017

  18. [26]

    Exact sampling with integer linear programs and random perturbations

    Kim, C., Sabharwal, A., and Ermon, S. Exact sampling with integer linear programs and random perturbations. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 30, 2016

  19. [27]

    Kingma, D. P. and Welling, M. Auto-encoding variational bayes. CoRR, abs/1312.6114, 2013

  20. [28]

    Learning multiple layers of features from tiny images

    Krizhevsky, A. Learning multiple layers of features from tiny images. Technical Report TR-2009, University of Toronto, 2009

  21. [29]

    Efficient dependency models: Simulating dependent random variables

    Lamboni, M. Efficient dependency models: Simulating dependent random variables. Mathematics and Computers in Simulation, 200: 0 199--217, 2022. ISSN 0378-4754. doi:https://doi.org/10.1016/j.matcom.2022.04.018

  22. [30]

    Liu, A., Niepert, M., and den Broeck, G. V. Image inpainting via tractable steering of diffusion models. In The Twelfth International Conference on Learning Representations, 2024

  23. [31]

    Deep learning face attributes in the wild

    Liu, Z., Luo, P., Wang, X., and Tang, X. Deep learning face attributes in the wild. In Proceedings of the IEEE International Conference on Computer Vision (ICCV), pp.\ 3730--3738, 2015

  24. [32]

    Dpm-solver: A fast ode solver for diffusion probabilistic model sampling in around 10 steps

    Lu, C., Zhou, Y., Bao, F., Chen, J., Li, C., and Zhu, J. Dpm-solver: A fast ode solver for diffusion probabilistic model sampling in around 10 steps. arXiv preprint arXiv:2206.00927, 2022

  25. [33]

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

    Maddison, C. J., Mnih, A., and Teh, Y. W. The concrete distribution: A continuous relaxation of discrete random variables. In International Conference on Learning Representations, 2017

  26. [34]

    Vael: Bridging variational autoencoders and probabilistic logic programming

    Misino, E., Marra, G., and Sansone, E. Vael: Bridging variational autoencoders and probabilistic logic programming. Advances in Neural Information Processing Systems, 35: 0 4667--4679, 2022

  27. [35]

    Raissi, M., Perdikaris, P., and Karniadakis, G. E. Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations. Journal of Computational Physics, 378: 0 686--707, 2019

  28. [36]

    M., and Fern, X

    Raza, A., Sturluson, A., Simon, C. M., and Fern, X. Message passing neural networks for partial charge assignment to metal--organic frameworks. The Journal of Physical Chemistry C, 124 0 (35): 0 19070--19082, 2020

  29. [37]

    Improved techniques for training gans

    Salimans, T., Goodfellow, I., Zaremba, W., Cheung, V., Radford, A., and Chen, X. Improved techniques for training gans. In Advances in Neural Information Processing Systems (NeurIPS), 2016

  30. [38]

    Sharpe, W. F. Mutual fund performance. Journal of Business, 39 0 (1): 0 119--138, 1966

  31. [39]

    A unified approach to count-based weakly supervised learning

    Shukla, V., Zeng, Z., Ahmed, K., and Van den Broeck, G. A unified approach to count-based weakly supervised learning. Advances in Neural Information Processing Systems, 36, 2024

  32. [40]

    K., Raiko, T., Maal e, L., S nderby, S

    S nderby, C. K., Raiko, T., Maal e, L., S nderby, S. r. K., and Winther, O. Ladder variational autoencoders. In Lee, D., Sugiyama, M., Luxburg, U., Guyon, I., and Garnett, R. (eds.), Advances in Neural Information Processing Systems, volume 29. Curran Associates, Inc., 2016

  33. [41]

    Denoising diffusion implicit models

    Song, J., Meng, C., and Ermon, S. Denoising diffusion implicit models. In International Conference on Learning Representations, 2021

  34. [42]

    and Ermon, S

    Song, Y. and Ermon, S. Generative modeling by estimating gradients of the data distribution. In Advances in Neural Information Processing Systems, pp.\ 11895--11907, 2019

  35. [43]

    C., Giunchiglia, E., and Lukasiewicz, T

    Stoian, M. C., Giunchiglia, E., and Lukasiewicz, T. Exploiting t-norms for deep learning in autonomous driving. In d'Avila Garcez, A. S., Besold, T. R., Gori, M., and Jiménez-Ruiz, E. (eds.), Proceedings of the 17th International Workshop on Neural-Symbolic Learning and Reason...

  36. [44]

    C., Dyrmishi, S., Cordy, M., Lukasiewicz, T., and Giunchiglia, E

    Stoian, M. C., Dyrmishi, S., Cordy, M., Lukasiewicz, T., and Giunchiglia, E. How realistic is your synthetic data? constraining deep generative models for tabular data. In The Twelfth International Conference on Learning Representations, 2024

  37. [45]

    Differentiable submodular maximization

    Tschiatschek, S., Sahin, A., and Krause, A. Differentiable submodular maximization. In International Joint Conference on Artificial Intelligence, 2018

  38. [46]

    Sampling the multivariate standard normal distribution under a weighted sum constraint

    Vrins, F. Sampling the multivariate standard normal distribution under a weighted sum constraint. Risks, 6 0 (3), 2018. ISSN 2227-9091. doi:10.3390/risks6030064

  39. [47]

    LinSATNet : The positive linear satisfiability neural networks

    Wang, R., Zhang, Y., Guo, Z., Chen, T., Yang, X., and Yan, J. LinSATNet : The positive linear satisfiability neural networks. In International Conference on Machine Learning (ICML), 2023

  40. [48]

    Melding the data-decisions pipeline: Decision-focused learning for combinatorial optimization

    Wilder, B. Melding the data-decisions pipeline: Decision-focused learning for combinatorial optimization. In Proceedings of the 33rd AAAI Conference on Artificial Intelligence, 2019

  41. [49]

    Xie, S. M. and Ermon, S. Reparameterizable subset sampling via continuous relaxations. International Joint Conference on Artificial Intelligence (IJCAI), 2019

  42. [50]

    A semantic loss function for deep learning with symbolic knowledge

    Xu, J., Zhang, Z., Friedman, T., Liang, Y., and Broeck, G. A semantic loss function for deep learning with symbolic knowledge. In International conference on machine learning, pp.\ 5502--5511. PMLR, 2018

  43. [51]

    Lsun: Construction of a large-scale image dataset using deep learning with humans in the loop

    Yu, F., Seff, A., Zhang, Y., Song, S., Funkhouser, T., and Xiao, J. Lsun: Construction of a large-scale image dataset using deep learning with humans in the loop. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, pp.\ 1--10, 2015

  44. [52]

    Physdiff: Physics-guided human motion diffusion model

    Yuan, Y., Song, J., Iqbal, U., Vahdat, A., and Kautz, J. Physdiff: Physics-guided human motion diffusion model. In Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), 2023

  45. [53]

    H., Meng, T., Chang, K.-W., and Van den Broeck, G

    Zhang, H., Li, L. H., Meng, T., Chang, K.-W., and Van den Broeck, G. On the paradox of learning to reason from data. In Proceedings of the Thirty-Second International Joint Conference on Artificial Intelligence, pp.\ 3365--3373, 2023

  46. [54]

    Deep learning for portfolio optimization

    Zhang, Z., Zohren, S., and Roberts, S. Deep learning for portfolio optimization. The Journal of Financial Data Science, 2 0 (4): 0 8--20, 2020

Pith tools

Reviewed August 8, 2026 · model on record in the stance chip above.