WWAE improves deep generative models by minimizing penalized optimal transport.
problem Challenges in learning deep generative models, such as blurriness in variational auto-encoders and instability in generative adversarial networks.
method WWAE formulates as minimization of penalized optimal transport between target and generated distributions. It uses closed-form squared Wasserstein-2 distance for Gaussians.
result WWAE learns better latent structures and generates higher quality samples than VAEs and GANs.
We propose the Wasserstein Auto-Encoder (WAE)---a new algorithm for building a generative model of the data distribution. WAE minimizes a penalized form of the Wasserstein distance between the model distribution and the target distribution, which leads to a different regularizer than the one used by the Variational Aut…
A new method uses persistent homology to assess auto-encoders' latent manifold quality.
problem Chaos in auto-encoders' latent manifold and failure of current distance measures.
method Persistent Homology for Wasserstein Auto-Encoders (PHom-WAE).
result PHom-WAE improves auto-encoders' performance in credit card transaction data.
Develops a new non-adversarial framework for better generative models.
problem Inaccurate approximation of target distribution in latent space.
method Tessellated Wasserstein Auto-Encoders (TWAE) using centroidal Voronoi tessellation (CVT) to tessellate latent space.
result Significantly enhances generative performance in terms of FID compared to existing models.
We study the role of latent space dimensionality in Wasserstein auto-encoders (WAEs). Through experimentation on synthetic and real datasets, we argue that random encoders should be preferred over deterministic encoders. We highlight the potential of WAEs for representation learning with promising results on a benchmar…
Generative model for morphological continuum of normal and pathological states.
problem Identifying trends and features that separate normality and pathology in biomedical images.
method Wasserstein Auto-encoder with HSIC regularization for latent features.
result Model generates a continuum of morphological changes corresponding to side information.
Generative source separation methods such as non-negative matrix factorization (NMF) or auto-encoders, rely on the assumption of an output probability density. Generative Adversarial Networks (GANs) can learn data distributions without needing a parametric assumption on the output density. We show on a speech source se…
DAG-WGAN learns causal structures using Wasserstein distance.
problem Learning causal structures from data with combinatorial challenges.
method Combines Wasserstein distance, auto-encoder, and acyclicity constraint.
result Demonstrates good performance compared to state-of-the-art models.
Generative models use DAE or DSM to estimate score, then Langevin sampling for sampling.
problem Estimating the score function of complex distributions for sampling.
method DAE or DSM for score estimation, Langevin sampling for sampling.
result Finite-sample bounds in Wasserstein distance for the sampling scheme.
Paper computes MMD for Gaussian kernel and proposes improvements for WAEs.
problem Estimating MMD between distributions for machine learning applications.
method Closed-form expressions for Gaussian kernel MMD, batch normalization at code layer.
result Improved interpretability and performance of WAEs with analytic MMD.
Proposes CWAE-IRL for efficient IRL in complex environments.
problem Learning reward functions from expert actions in MDPs.
method Variational inference with CVAE and Wasserstein loss.
result CWAE-IRL learns reward functions efficiently in complex environments.
Fisher auto-encoders use Fisher divergence for more robust generative modeling.
problem Model uncertainty in generative models.
method Minimizing Fisher divergence between true and modeled joint distributions.
result Fisher auto-encoders can more accurately quantify model uncertainty.
Paper presents a new method to learn unnormalized models efficiently.
problem Scalability issues in score matching for flexible unnormalized models.
method Connects learning objectives to Wasserstein gradient flows for scalability.
result Demonstrates improved learning of unnormalized models on manifolds.
We study unsupervised generative modeling in terms of the optimal transport (OT) problem between true (but unknown) data distribution PX and the latent variable model distribution PG. We show that the OT problem can be equivalently written in terms of probabilistic encoders, which are constrained to match the pos…
A new spherical Sliced-Wasserstein distance for data on spheres.
problem Defining Wasserstein distance on manifolds, especially spheres.
method Closed-form solutions of the Wasserstein distance on the circle and a new spherical Radon transform.
result A novel spherical Sliced-Wasserstein (SW) discrepancy for data on spheres.
Max-sliced Wasserstein distance improves GAN training on high-dimensional images.
problem Improving GAN training on high-dimensional distributions with reduced hyper-parameters and training time.
method Developed max-sliced Wasserstein distance to reduce projection complexity and sample complexity.
result Trains GANs on high-dimensional images up to 256x256 resolution.
Adaptive framework for learning latent space dimensions in GANs.
problem Inadequate latent space dimensions lead to poor generative models for complex data.
method Proposes a novel framework (LWGAN) that adaptively learns latent dimensions of data manifolds.
result Proves that the estimated intrinsic dimension is a consistent estimate of the true data manifold dimension.
Paper defines new GSW distances for probability measures.
problem Computational simplicity and similarity to Wasserstein distance.
method Generalized Radon transform to define GSW distances.
result GSW and max-GSW distances are distances under certain conditions.
A new method for estimating complex models and high-dimensional data.
problem Difficulty in computing Hessian of log-density functions for complex models and high-dimensional data.
method Sliced score matching, which projects scores onto random vectors before comparison.
result Sliced score matching can learn deep energy-based models and produce accurate score estimates.
New DG method minimizes barycentric alignment and reconstruction loss.
problem Improving domain generalization in machine learning.
method Introduces a new upper bound and WBAE algorithm.
result WBAE outperforms state-of-the-art DG algorithms.
LANCA uses ANM to learn latent causal factors without supervision.
problem Learning latent causal factors without supervision.
method LANCA employs a deterministic Wasserstein Auto-Encoder coupled with a differentiable ANM Layer.
result LANCA outperforms baselines on physics and photorealistic environments.
FlexAE addresses bias-variance trade-off in RAEs by learning latent priors.
problem Improving generation quality of deterministic AE models.
method Introducing flexibly learnable latent priors in WAEs to optimize the latent distribution.
result FlexAE achieves state-of-the-art performance in AE-based generative models.
PHom-GeM uses topological features to assess generative models.
problem Generative models produce chaotic distributions during training.
method Persistent Homology for Generative Models (PHom-GeM) minimizes an objective function between true and reconstructed distributions.
result PHom-GeM is a topological distance measure for generative models.
iWGAN improves GANs by stabilizing training and preventing mode collapse.
problem Stable and effective training of GANs with mode collapse.
method iWGAN combines auto-encoders and WGANs using iterative primal dual optimization.
result iWGAN provides a clear stopping criterion and mitigates mode collapse.
We introduce Primal-Dual Wasserstein GAN, a new learning algorithm for building latent variable models of the data distribution based on the primal and the dual formulations of the optimal transport (OT) problem. We utilize the primal formulation to learn a flexible inference mechanism and to create an optimal approxim…
In generative modeling, the Wasserstein distance (WD) has emerged as a useful metric to measure the discrepancy between generated and real data distributions. Unfortunately, it is challenging to approximate the WD of high-dimensional distributions. In contrast, the sliced Wasserstein distance (SWD) factorizes high-dime…
We propose a new equilibrium enforcing method paired with a loss derived from the Wasserstein distance for training auto-encoder based Generative Adversarial Networks. This method balances the generator and discriminator during training. Additionally, it provides a new approximate convergence measure, fast and stable t…
Novel framework evaluates deep learning models under distribution shift.
problem Evaluating robustness and generalization of deep learning models under distribution shift.
method Variational Inference, Auto-Encoding Variational Bayes, Variational Gaussian Mixture Model, Wasserstein distance.
result Popular CNN architectures and Bayesian CNN models robustness and generalization under distribution shift are assessed.
Optimal transport offers an alternative to maximum likelihood for learning generative autoencoding models. We show that minimizing the p-Wasserstein distance between the generator and the true data distribution is equivalent to the unconstrained min-min optimization of the p-Wasserstein distance between the encoder agg…
New autoencoder uses goodness-of-fit tests for better model performance.
problem Improving the goodness-of-fit in generative models.
method Develops Goodness-of-Fit Autoencoder (GoFAE) incorporating GoF tests at minibatch and global levels.
result GoFAE achieves comparable performance to deep generative models while retaining statistical indistinguishability.
The paper identifies saddlepoints in unsupervised auto-encoding neural nets.
problem The risk landscape of unsupervised least squares in auto-encoding neural nets.
method Established an equivalence between unsupervised least squares and principal manifolds, discussed regularization strategies for auto-encoders.
result All non-trivial critical points in auto-encoding are saddlepoints, which are degenerate in overcomplete auto-encoding.
Two graph auto-encoders decouple feature propagation from graph convolution layers.
problem Designing efficient graph auto-encoders with fixed receptive fields.
method L-GAE and L-VGAE using linear matrix computation before auto-encoder input.
result Comparable performance to VGAEs with smaller, simpler networks.
New auto-encoder handles varying noise levels without retraining.
problem Auto-encoders degrade in noisy conditions.
method Formalized auto-encoders as transform learning, derived new architecture.
result Models generalize well to different noise levels.
A novel cross-modal auto-encoder associates different data types efficiently.
problem Cross-modal data association in heterogeneous datasets.
method Bayesian inference framework with variational auto-encoders and associators.
result Successfully associates visual and auditory data with minimal paired data.
DGA and DVGA learn disentangled graph representations to improve graph analysis.
problem Holistic graph auto-encoders fail to capture latent factors effectively.
method Design disentangled graph convolutional network and component-wise flow, impose independence constraints.
result Improved disentangled graph representations enhance graph analysis tasks.
A new triad decoder improves graph auto-encoders' performance.
problem Graph auto-encoders ignore edge interactions, leading to suboptimal predictions.
method Integrates triadic closure property to predict three edges in a local triad.
result Triad decoder leads to more accurate predictions, clustering, and graph characteristics preservation.
Recent work suggests that some auto-encoder variants do a good job of capturing the local manifold structure of the unknown data generating density. This paper contributes to the mathematical understanding of this phenomenon and helps define better justified sampling algorithms for deep learning based on auto-encoder v…
Graph attention auto-encoder reconstructs graph structure and attributes.
problem Lack of methods to reconstruct graph structure and node attributes in graph auto-encoders.
method Stacked encoder/decoder layers with self-attention mechanisms, regularized node representations to reconstruct graph structure.
result Competitive performance on node classification benchmarks, including inductive learning.
What do auto-encoders learn about the underlying data generating distribution? Recent work suggests that some auto-encoder variants do a good job of capturing the local manifold structure of data. This paper clarifies some of these previous observations by showing that minimizing a particular form of regularized recons…
Paper proposes using auto-encoders to efficiently purify adversarial perturbations.
problem Vulnerability of machine learning models to adversarial examples.
method Iterative adversarial training on an auto-encoder to purify perturbations.
result The auto-encoder trained model outperforms other methods in protecting models against white-box attacks.
CVAEs improve VAEs by accounting for correlations in latent representations.
problem VAEs fail to account for correlations between data points, limiting their effectiveness.
method CVAEs incorporate correlation structure into VAEs using a prior and tractable approximations.
result CVAEs outperform baseline algorithms in matching and link prediction tasks.
Paper proposes a novel auto-encoder for latent density estimation.
problem Challenges of learning generative probabilistic models due to curse of dimensionality.
method Joint dimensionality reduction and non-parametric density estimation framework using a novel estimator.
result Proposed model achieves promising results on various datasets.
New distribution on simplex for auto-encoding tasks.
problem Developing a new distribution for sparsity in auto-encoding models.
method Using Kumaraswamy distribution and ordered stick-breaking process.
result The new distribution has an exact and closed form reparameterization.
Principal components analysis (PCA) is the optimal linear auto-encoder of data, and it is often used to construct features. Enforcing sparsity on the principal components can promote better generalization, while improving the interpretability of the features. We study the problem of constructing optimal sparse linear a…
CasVAE outperforms supervised methods for star-galaxy classification.
problem Challenges in machine learning for astronomy data.
method Cascade Variational Auto-Encoder (CasVAE) for unsupervised star-galaxy classification.
result CasVAE outperforms baseline models in accuracy and stability.
Auto-encoding generative adversarial networks (GANs) combine the standard GAN algorithm, which discriminates between real and model-generated data, with a reconstruction loss given by an auto-encoder. Such models aim to prevent mode collapse in the learned generative model by ensuring that it is grounded in all the ava…
SIG-VAE enhances VGAE for graph data modeling.
problem Limited flexibility in VGAE for graph data.
method Hierarchical variational framework with Bernoulli-Poisson link decoder.
result SIG-VAE outperforms state-of-the-art methods on graph tasks.
Paper proposes a new method for learning deep generative models using annealed importance sampling.
problem Learning deep generative models efficiently and accurately.
method Proposes annealed importance sampling as a bridge between variational inference and Markov chain Monte Carlo.
result Demonstrates improved density models and better trade-off between computation and model accuracy.