Wasserstein
PulseAugur coverage of Wasserstein — every cluster mentioning Wasserstein across labs, papers, and developer communities, ranked by signal.
12 day(s) with sentiment data
-
New framework tackles risk-averse online learning challenges
Researchers have developed a new framework for risk-averse online learning that addresses challenges in convergence. This approach models the problem as an online saddle-point stochastic game, where a decision-maker and…
-
New Wasserstein Policy Gradient Method for LQ Control Problems
Researchers have developed a Wasserstein policy gradient (WPG) method for entropy-regularized linear-quadratic (LQ) control problems. This approach leverages the fact that unrestricted LQ control problems have linear-Ga…
-
New sub-quadratic method improves bisimulation metric computation for MDPs
Researchers have developed a novel sub-quadratic method for calculating bisimulation metrics in Markov decision processes (MDPs). This new approach utilizes approximate nearest neighbor (ANN) indexing to efficiently sel…
-
Deep learning particle systems analyzed for convergence in neural SDEs
This paper analyzes controlled particle systems that arise in deep learning, specifically focusing on neural stochastic differential equations (SDEs). Researchers investigated the limiting behavior of sampled optimal co…
-
New research improves Wasserstein mixing time estimates for Langevin algorithm
A new paper published on arXiv details improved estimates for the unadjusted Langevin algorithm's asymptotic bias. The research provides a new bound for Wasserstein mixing time, achieving an order of $\kappa \sqrt{d}/\v…
-
New research explores Wasserstein gradient flows for Maximum Mean Discrepancy
Researchers have published a paper detailing Wasserstein gradient flows for Maximum Mean Discrepancy (MMD) using energy kernels. The study addresses challenges in applying standard gradient flow theory to nonsmooth kern…
-
New Voronoi Histogram Method Enhances Topological Data Analysis
Researchers have developed a new method called Voronoi histograms for vectorizing Expected Persistence Diagrams (EPDs), which are used to analyze the topology of point cloud data. This approach offers an alternative to …
-
Distributional random forests advanced for complex data and survey designs · 2 sources tracked
Two new research papers explore advanced applications of distributional random forests, moving beyond traditional mean-based splitting. The first paper introduces extensions to distributional splitting criteria, includi…
-
New Wasserstein Barycenter Solver Achieves State-of-the-Art Performance
Researchers have developed a new method for computing Wasserstein barycenters, which are used to aggregate probability measures while preserving geometric properties. This novel approach utilizes gradient flows in the s…
-
New Entropic Curvature Method Enhances Graph Neural Networks
Researchers have introduced a novel concept called Entropic Curvature to address limitations in Graph Neural Networks (GNNs), specifically the issues of oversmoothing and oversquashing. This new approach extends existin…
-
New distributional clustering method introduced using Wasserstein kernel
Researchers have introduced the distributional determinantal point process (dDPP), a novel method for clustering probability distributions. This dDPP utilizes a sliced Wasserstein kernel and is demonstrated to be a vali…
-
New math paper details optimization on Wasserstein space
Researchers have developed new methods for optimizing functionals on Wasserstein spaces, a mathematical concept crucial for understanding probability distributions. The work establishes linear convergence for proximal d…
-
New research links larger LLM capacity to improved safety under adversarial shifts
A new research paper introduces a robust meta-learning framework for in-context learning (ICL) in large language models, specifically focusing on Transformers. The framework provides theoretical guarantees against adver…
-
New framework unifies vector quantization for improved visual representation learning
Researchers have developed a new framework for vector quantization (VQ) that addresses common issues like training instability and codebook collapse. The proposed distributional matching framework aims to align the dist…
-
New optimal self-distillation method improves generative model training
Researchers have developed a method called optimal self-distillation (SD) for rectified flow (RF) models, aiming to improve generative model training. This technique involves training a student model on a mix of true RF…
-
New DRO method adapts ambiguity sets for better financial optimization
Researchers have developed a new method called Learned Predictive Ambiguity Sets (LPAS) for decision-focused distributionally robust optimization. Unlike traditional methods that use fixed ambiguity sets, LPAS employs a…
-
New framework PeTeR enhances probabilistic circuit robustness post-training
Researchers have introduced PeTeR, a new post-training framework designed to enhance the robustness of probabilistic circuits (PCs) against distribution shifts. Unlike existing methods that require training from scratch…
-
New research demystifies deep ReLU networks and SGD training dynamics
Two new research papers explore the underlying principles and training dynamics of deep feedforward ReLU networks. The first paper delves into the mechanism of these networks, explaining how hidden layer units create pi…
-
New 'stitching' method reconstructs population dynamics via Wasserstein residuals
Researchers have developed a new particle-based method called 'stitching' for reconstructing population dynamics, which are often modeled as Wasserstein gradient flows. This novel approach bypasses costly optimal transp…
-
New research explores Wasserstein DRO for risk-sensitive estimation and regret optimization
Two new research papers explore the application of Wasserstein distributionally robust optimization (DRO) in different machine learning contexts. The first paper introduces a method for risk-sensitive estimation using W…