Conformal mirror descent with logarithmic divergences

DOI10.1007/S41884-022-00089-3arXiv2209.02938MaRDI QIDQ6138802FDOQ6138802

Authors: Amanjit Singh Kainth, Ting-Kam Leonard Wong, Frank Rudzicz 

Publication date: 16 January 2024

Published in: Information Geometry (Search for Journal in Brave)

Abstract: The logarithmic divergence is an extension of the Bregman divergence motivated by optimal transport and a generalized convex duality, and satisfies many remarkable properties. Using the geometry induced by the logarithmic divergence, we introduce a generalization of continuous time mirror descent that we term the conformal mirror descent. We derive its dynamics under a generalized mirror map, and show that it is a time change of a corresponding Hessian gradient flow. We also prove convergence results in continuous time. We apply the conformal mirror descent to online estimation of a generalized exponential family, and construct a family of gradient flows on the unit simplex via the Dirichlet optimal transport problem.

Full work available at URL: https://arxiv.org/abs/2209.02938

Recommendations

• Logarithmic divergences: geometry and interpretation of curvature
• The information geometry of mirror descent
• Convergence of online mirror descent
• Logarithmic divergences from optimal transport and Rényi geometry
• Hessian informed mirror descent

zbMATH Keywords

logarithmic divergenceconformal Hessian metricconformal mirror descentDirichlet optimal transportHessian gradient flow

Mathematics Subject Classification ID

Calculus of variations and optimal control; optimization (49-XX)

Cites Work

• A Relationship Between Arbitrary Positive Matrices and Doubly Stochastic Matrices
• Isometric logratio transformations for compositional data analysis
• Polar factorization and monotone rearrangement of vector‐valued functions
• Clustering with Bregman divergences.
• Title not available (Why is that?)
• Mirror descent and nonlinear projected subgradient methods for convex optimization.
• Title not available (Why is that?)
• The geometry of Hessian structures
• Title not available (Why is that?)
• Generalised Thermostatistics
• Title not available (Why is that?)
• Geometry of minimum contrast
• Information Geometry of U-Boost and Bregman Divergence
• Logistic regression, AdaBoost and Bregman distances
• Rényi Divergence and Kullback-Leibler Divergence
• Continuity, curvature, and the general covariance of optimal transportation
• Stability of a 4th-order curvature condition arising in optimal transport theory
• The geometry of relative arbitrage
• Title not available (Why is that?)
• Hessian Riemannian Gradient Flows in Convex Programming
• Gradient systems in view of information geometry
• Information geometry
• Information geometry and its applications
• Logarithmic divergences: geometry and interpretation of curvature
• Exponentially concave functions and a new information geometry
• Information geometry
• A variational perspective on accelerated methods in optimization
• Information geometry in portfolio theory
• Logarithmic divergences from optimal transport and Rényi geometry
• Multiplicative Schrödinger problem and the Dirichlet transport
• The Information Geometry of Mirror Descent
• A regression model for compositional data based on the shifted-Dirichlet distribution
• Cramér-Rao lower bounds arising from generalized Csiszár divergences
• Projection theorems and estimating equations for power-law models
• On Conformal Divergences and Their Population Minimizers
• Pseudo-Riemannian geometry encodes information geometry in optimal transport
• A gradient descent perspective on Sinkhorn
• Minimum Divergence Methods in Statistical Machine Learning
• Tsallis and Rényi Deformations Linked via a New λ-Duality

Cited In (1)

• On variational inference and maximum likelihood estimation with the \(\lambda\)-exponential family

This page was built for publication: Conformal mirror descent with logarithmic divergences

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q6138802)