Paper relaxes the Lipschitz constraint in WGANs to improve performance.
problem WGANs do not always outperform other GAN variants due to imperfect implementation of the Lipschitz condition.
method Proposes a new dual form of Wasserstein distance (Sobolev duality) that relaxes the Lipschitz constraint but maintains gradient property.
result SWGAN, based on Sobolev duality, outperforms existing methods in experiments.
We propose an SDP relaxation for the Gromov-Wasserstein distance, providing globally optimal solutions.
problem Matching objects between incomparable spaces using the Gromov-Wasserstein distance.
method Semi-definite programming (SDP) relaxation of the GW distance.
result The SDP relaxation provides globally optimal solutions for the GW distance in some instances.
Wasserstein Generative Adversarial Networks (WGANs) provide a versatile class of models, which have attracted great attention in various applications. However, this framework has two main drawbacks: (i) Wasserstein-1 (or Earth-Mover) distance is restrictive such that WGANs cannot always fit data geometry well; (ii) It …
Optimal neural network approximation for Wasserstein gradient direction via convex optimization.
problem Approximating Wasserstein gradient direction with limited data.
method Two-layer networks with squared-ReLU activations, SDP relaxation.
result Optimal approximation of Wasserstein gradient direction in two-layer networks.
Paper proposes a new Wasserstein distance for mixtures of radially contoured distributions.
problem Generalization of Wasserstein distance to non-elliptically contoured distributions.
method Relaxed formulation for mixtures of radially contoured distributions without marginal consistency.
result The new distance yields more stable error and better color distribution in image transfer tasks.
A new Wasserstein K-means method for clustering probability distributions.
problem Clustering probability distributions using the Wasserstein metric.
method Distance-based K-means with SDP relaxation for Wasserstein barycenters. result Distance-based K-means outperforms centroid-based K-means for clustering probability distributions. Optimal transport relaxations improve Wasserstein GAN training efficiency.
problem Training inefficiency in Wasserstein GANs.
method Optimal transport relaxations with a minimization step over a small region.
result Running time improvements with no performance degradation.
New algorithm for low-rank optimal transport with improved interpretability and efficiency.
problem Quadratic scaling of optimal transport coupling matrix for massive datasets.
method Factor Relaxation with Latent Coupling (FRLC) algorithm.
result Superior performance on diverse applications including graph clustering and spatial transcriptomics.
Paper develops KMS Wasserstein for high-dimensional data reduction.
problem Optimal transport's curse of dimensionality in high-dimensional data.
method Kernel max-sliced (KMS) Wasserstein distance for dimensionality reduction.
result Sharp finite-sample guarantees for KMS p-Wasserstein distance. Paper proposes PRWB and RPRWB for Wasserstein barycenters.
problem Numerical challenges in computing Wasserstein barycenters.
method Projection robust Wasserstein barycenter (PRWB) and relaxed PRWB (RPRWB).
result RPRWB improves clustering performance on real text datasets.
Adapts DR objectives for both sample and feature size reduction.
problem Simultaneously reduce sample and feature sizes.
method Semi-relaxed Gromov-Wasserstein optimal transport.
result OT plan delivers competitive hard clustering.
New formulations for comparing metric measure spaces with arbitrary positive measures.
problem Comparing metric measure spaces with arbitrary positive measures.
method Two novel formulations: a divergence and a conic lifting approach.
result Efficiently solvable formulations for comparing metric spaces with arbitrary positive measures.
Proposes a robust variant of Wasserstein distance using subspace projections.
problem Challenges in computing Wasserstein distances in high dimensions.
method Max-min robust variant, orthogonal projections on subspaces, convex relaxation.
result Robust Wasserstein distances inherit favorable properties from OT geometry.
Paper proposes models to identify topic sparsity in social media texts.
problem Topic sparsity in online social media content.
method Sparsemax and relaxed Wasserstein for topic sparsity.
result Proposed models achieve better topic sparsity identification and performance.
Paper proposes a new method for predicting drug interactions using adversarial autoencoders.
problem Predicting drug interactions to prevent adverse events.
method Introduces adversarial autoencoders based on Wasserstein distances and Gumbel-Softmax relaxation to generate high-quality negative samples.
result Significant improvements in link prediction and DDI classification tasks.
Unified framework for fair regression in aware and unaware settings.
problem Lack of principled methods for fair regression in unawareness settings.
method Formulated as an optimal transport problem, unifying aware and unaware settings.
result Characterizes optimal prediction functions via optimal transport maps under different penalties.
Method learns graphons from graphs via Gromov-Wasserstein barycenters.
problem Learning nonparametric graph models from finite graphs.
method Approximate graphons with step functions, use Gromov-Wasserstein distance, learn barycenters.
result Proposed method outperforms state-of-the-art on synthetic and real-world data.
HMC achieves optimal convergence rate for strongly logconcave distributions.
problem Sampling from strongly logconcave densities efficiently.
method Hamiltonian Monte Carlo (HMC) with an optimal ODE solver.
result HMC achieves an optimal convergence rate of O(κ) for sampling from strongly logconcave distributions. Paper analyzes SGLD for nonconvex optimization with local conditions.
problem Analyzing sampling algorithms for nonconvex optimization.
method Non-asymptotic estimates for SGLD under local conditions.
result Establishes error bounds for expected excess risk.
Paper bridges f-GANs and WGANs for better image generation.
problem Learning high-dimensional distributions using GANs.
method List constraints, minimize Lagrangian relaxation, propose KL-Wasserstein GAN.
result Empirical success on synthetic and real-world image generation benchmarks.
Paper presents efficient computation of robust Wasserstein distance using Riemannian optimization.
problem Intractability of optimizing Projection Robust Wasserstein (PRW) distance due to non-convexity and non-smoothness.
method Riemannian optimization to efficiently compute PRW/Wasserstein Projection Pursuit (WPP) distance.
result The original formulation of PRW/WPP can be efficiently computed in practice, providing better behavior than its convex relaxation.
This paper tackles robust control of noisy systems with uncertain distributions.
problem Optimal control of sampled-data stochastic systems with multiplicative noise and distributional ambiguity.
method Develops a convex relaxation to handle the ``concave-max'' geometry and derives a probabilistic performance guarantee.
result Derives an explicit, non-asymptotic bound on the duality gap and proves robust viability conditions.
The paper explores arbitrage in financial markets under uncertainty using Wasserstein distance.
problem Investigating arbitrage in financial markets with distributional uncertainty.
method Using Wasserstein distance, the paper considers weak and strong forms of arbitrage conditions and introduces a relaxation called statistical arbitrage.
result The paper derives dual formulations of robust arbitrage conditions and conducts computational experiments to answer questions about ambiguity and statistical arbitrage.
Improved robustness in multivariate regression and classification with DRO under Wasserstein metric.
problem Outliers in covariates and responses.
method Distributionally Robust Optimization (DRO) with Wasserstein metric ambiguity set and regularization.
result Significant improvement in predictive error and robustness.
A new method for conditional sampling using paired Wasserstein Autoencoders.
problem Conditional sampling from complex data distributions.
method Derive a novel loss function for Wasserstein Autoencoders to enable sampling from OT-type couplings.
result Learned cost-optimal transport maps and conditional sampling from an OT-type coupling.
A new framework improves fairness in clustering and Wasserstein Barycenter problems.
problem Fair clustering in datasets with multiple groups.
method Relax and Merge framework for (1+4ρ+O(ε))-approximate solutions. result Improved approximation guarantees for fairness constraints.
Proposes a new graph kernel framework using regularized Wasserstein distances.
problem Learning optimal transport distances for graph kernels.
method Introduces Regularized Wasserstein (RW) discrepancy with two regularization terms.
result Empirically validated method outperforms state-of-the-art methods.
Paper tackles distribution matching by partially matching distributions, achieving robust results.
problem Robustly aligning two probability distributions.
method Developed a partial Wasserstein adversarial network (PWAN) to efficiently approximate the partial Wasserstein-1 (PW) discrepancy.
result The PWAN effectively produces highly robust matching results, outperforming state-of-the-art methods.
Regularization helps protect machine learning models from poisoning attacks.
problem Mitigating the impact of poisoned data on machine learning models.
method Distributionally-robust optimization using Wasserstein distance to find an upper bound for worst-case fitness.
result The regularizer is equal to the dual norm of the model parameters for regression models.
We introduce principal differences analysis (PDA) for analyzing differences between high-dimensional distributions. The method operates by finding the projection that maximizes the Wasserstein divergence between the resulting univariate populations. Relying on the Cramer-Wold device, it requires no assumptions about th…
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…
Improved spectral clustering via Gromov-Wasserstein Learning.
problem Optimizing graph partitioning performance.
method Bridge spectral clustering and GWL, using heat kernel for stable node correspondences.
result Improved graph partitioning results without compromising theoretical guarantees.
A new method optimizes slicing directions for SW distances to improve high-dimensional probability measure comparison.
problem Challenging identification of informative slicing directions for SW distances.
method Constrained learning approach to optimize slicing directions, using continuous relaxations and gradient-based primal-dual approach.
result Demonstrated efficacy in learning more informative slicing directions on various high-dimensional data.
Develops robust learning framework under distributional perturbations.
problem Learning robust to data distributional changes.
method Distributionally Robust Optimization (DRO) under Wasserstein metric.
result Establishes performance guarantees and tractable formulations.
Given a family of probability measures in P(X), the space of probability measures on a Hilbert space X, our goal in this paper is to highlight one ore more curves in P(X) that summarize efficiently that family. We propose to study this problem under the optimal transport (Wasserstein) geometry, using curves that are re…
Paper proposes an algorithm for sampling from complex mixture distributions without requiring smoothness.
problem Sampling from a mixture of weakly smooth potentials.
method Unadjusted Langevin algorithm with Euler discretization for a mixture of weakly smooth distributions.
result Convergence in Kullback-Leibler divergence and Lβ-Wasserstein metric with polynomial dependence on dimension. This paper introduces a new formulation of the Conic Gromov-Wasserstein distance for comparing complex network structures.
problem Comparing measures of unequal mass and complex network structures.
method Novel semi-coupling formulation and extension to hypernetworks.
result Establishes fundamental properties and robustness of CGW metric.
Lipschitz GANs solve gradient uninformativeness in GANs.
problem Gradient uninformativeness in GANs.
method Introduce Lipschitz constraint on the discriminative function space.
result Lipschitz GANs eliminate gradient uninformativeness and generate better quality samples.
Optimizes decisions in time-varying distributions using online stochastic methods and Wasserstein distance.
problem Optimizing decisions in time-varying distributions using Wasserstein distance.
method Online proximal-gradient method, exact penalty method, constraint-tightening approach.
result Dynamic regret bounds for tracking and estimation error.
New algorithm improves clustering and quantization using MMD.
problem Approximating probability distributions with weighted mixtures of Dirac measures.
method Gradient flow, mean shift, and MMD-optimal quantization.
result MSIP algorithm is more robust than state-of-the-art methods.
This work improves understanding of dimension reduction algorithms and their probabilistic embeddings.
problem Improving theoretical understanding of non-linear dimension reduction algorithms.
method Analytical investigation of a generalized multidimensional scaling optimization problem.
result Probabilistic formulation of the problem leads to deterministic embeddings, contrary to standard implementations.
Study matches two noisy point clouds with geometric transformations and relabeling.
problem Matching two noisy point clouds with orthogonal transformations and relabeling.
method Information-theoretic results and Ping-Pong algorithm for computational alignment.
result The Ping-Pong algorithm retrieves the planted signal after one step.
In this paper we study the frequentist convergence rate for the Latent Dirichlet Allocation (Blei et al., 2003) topic models. We show that the maximum likelihood estimator converges to one of the finitely many equivalent parameters in Wasserstein's distance metric at a rate of n−1/4 without assuming separability o…
New method improves robust point matching under probabilistic settings.
problem Insufficient theoretical understanding of existing point matching methods.
method Distance profiles and modified matching procedure.
result Improved robustness under probabilistic settings.
Develops a robust multiclass classification method for deep image classifiers.
problem Tackles data contamination and robustness to outliers in deep image classifiers.
method Uses Distributionally Robust Optimization (DRO) with Wasserstein metric ambiguity sets and regularized learning.
result Reduces test error rate by up to 83.5% and loss by up to 91.3% in image classification tasks.
FairWASP optimizes training data to reduce disparities across subgroups.
problem Reducing disparities in model outputs across different subgroups in machine learning.
method A novel pre-processing approach that minimizes Wasserstein distance to the original dataset while satisfying demographic parity.
result Integer weights are optimal, allowing FairWASP to be understood as duplicating or eliminating samples.
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.
We present a Distributionally Robust Optimization (DRO) approach to estimate a robustified regression plane in a linear regression setting, when the observed samples are potentially contaminated with adversarially corrupted outliers. Our approach mitigates the impact of outliers through hedging against a family of dist…