Approximation Based Variance Reduction for Reparameterization Gradients

Abstract: Flexible variational distributions improve variational inference but are harder to optimize. In this work we present a control variate that is applicable for any reparameterizable distribution with known mean and covariance, e.g. Gaussians with any covariance structure. The control variate is based on a quadratic approximation of the model, and its parameters are set using a double-descent scheme. We empirically show that this control variate leads to large improvements in gradient variance and optimization convergence for inference with non-factorized variational distributions.

13/04/2021

Machine Learning, Learning Generative Models, Time-series; Data Streams, Unsupervised Learning, Approximate Probabilistic Inference

13:39

13/04/2021

Approximation Based Variance Reduction for Reparameterization Gradients

Tomas Geffner, Justin Domke

Comments

Similar Papers

Nearest neighbour based estimates of gradients: Sharp nonasymptotic bounds and applications

Guillaume Ausset, Stephan Clémencon, François Portier

Keywords Abstract Paper

Walsh-Hadamard Variational Inference for Bayesian Deep Learning

Simone Rossi, Sebastien Marmin, Maurizio Filippone

Keywords Abstract Paper

Interior Point Solving for LP-based prediction+optimisation

Jayanta Mandi, Tias Guns

Keywords Abstract Paper

Monte Carlo Filtering Objectives

Shuangshuang Chen, Sihao Ding, Yiannis Karayiannidis, Mårten Björkman

Keywords Abstract Paper

Machine Learning, Learning Generative Models, Time-series; Data Streams, Unsupervised Learning, Approximate Probabilistic Inference

Direct loss minimization for sparse gaussian processes

Yadi Wei, Rishit Sheth, Roni Khardon

Keywords Abstract Paper

Stochastic Optimization for Regularized Wasserstein Estimators

Marin Ballu, Quentin Berthet, Francis Bach

Keywords Abstract Paper

Robust Regression Revisited: Acceleration and Improved Estimation Rates

Arun Jambulapati, Jerry Li, Tselil Schramm, Kevin Tian

Keywords Abstract Paper

theory, optimization

Hessian Eigenspectra of More Realistic Nonlinear Models

Zhenyu Liao, Michael W Mahoney

Keywords Abstract Paper

theory, optimization, machine learning

Handling the Positive-Definite Constraint in the Bayesian Learning Rule

Wu Lin, Mark Schmidt, Mohammad Emtiyaz Khan

Keywords Abstract Paper

Probabilistic Inference - Approximate, Monte Carlo, and Spectral Methods

Variational Model-based Policy Optimization

Yinlam Chow, Brandon Cui, Moonkyung Ryu, Mohammad Ghavamzadeh

Keywords Abstract Paper

Machine Learning, Reinforcement Learning

Learning Hierarchical Interactions at Scale: A Convex Optimization Approach

Hussein Hazimeh, Rahul Mazumder

Keywords Abstract Paper

Marginalized Stochastic Natural Gradients for Black-Box Variational Inference

Geng Ji, Debora Sujono, Erik Sudderth

Keywords Abstract Paper

Probabilistic Methods, Approximate Inference

Distributional Gradient Matching for Learning Uncertain Neural Dynamics Models

Lenart Treven, Philippe Wenk, Florian Dorfler, Andreas Krause

Keywords Abstract Paper

deep learning, reinforcement learning and planning, kernel methods, active learning

Continuous Latent Process Flows

Ruizhi Deng, Marcus Brubaker, Greg Mori, Andreas M Lehrmann

Keywords Abstract Paper

Direct Policy Gradients: Direct Optimization of Policies in Discrete Action Spaces

Guy Lorberbom, Chris J. Maddison, Nicolas Heess and Tamir Hazan, Daniel Tarlow

Keywords Abstract Paper

Improved CP-Based Lagrangian Relaxation Approach with an Application to the TSP

Raphaël Boudreault, Claude-Guy Quimper

Keywords Abstract Paper

Constraints and SAT, Global Constraints, Constraint Optimization, Constraints

Beyond Linearization: On Quadratic and Higher-Order Approximation of Wide Neural Networks

Yu Bai, Jason D. Lee

Keywords Abstract Paper

Neural Tangent Kernels, over-parametrized neural networks, deep learning theory

A convex optimization formulation for multivariate regression

Keywords Abstract Paper

Beyond the Mean-Field: Structured Deep Gaussian Processes Improve the Predictive Uncertainties

Jakob Lindinger, David Reeb, Christoph Lippert, Barbara Rakitsch

Keywords Abstract Paper

Multivariate Probabilistic Time Series Forecasting via Conditioned Normalizing Flows

Kashif Rasul, Abdul-Saboor Sheikh, Ingmar Schuster and Urs Bergmann, Roland Vollgraf

Keywords Abstract Paper

probabilistic multivariate forecasting, normalizing flows, attention, time series

Non-convex Distributionally Robust Optimization: Non-asymptotic Analysis

Jikai Jin, Bohang Zhang, Haiyang Wang, Liwei Wang

Keywords Abstract Paper

On Estimation in Latent Variable Models

Guanhua Fang, Ping Li

Keywords Abstract Paper

Optimization, Stochastic Optimization

Keywords Paper

Keywords Paper

Keywords Paper

Keywords Paper

Keywords Paper

Keywords Paper

Keywords Paper

Keywords Paper

Keywords Paper

Keywords Paper

Keywords Paper

Keywords Paper

Keywords Paper

Keywords Paper

Guy Lorberbom, Chris J. Maddison, Nicolas Heess and
Tamir Hazan, Daniel Tarlow

Keywords Paper

Keywords Paper

Keywords Paper

Keywords Paper

Keywords Paper

Kashif Rasul, Abdul-Saboor Sheikh, Ingmar Schuster and
Urs Bergmann, Roland Vollgraf

Keywords Paper

Keywords Paper

Keywords Paper

Ba-Hien Tran, Simone Rossi, Dimitrios Milios and
Pietro Michiardi, Edwin Bonilla, Maurizio Filippone

Keywords Paper

Keywords Paper

Keywords Paper

Keywords Paper

Keywords Paper

Keywords Paper

Keywords Paper

Jincheng Mei, Yue Gao, Bo Dai and
Csaba Szepesvari, Dale Schuurmans

Keywords Paper

Keywords Paper

Ali Hashemi, Yijing Gao, Chang Cai and
Sanjay Ghosh, Klaus-Robert Müller, Srikantan Nagarajan, Stefan Haufe

Keywords Paper

Shun-ichi Amari, Jimmy Ba, Roger Grosse and
Chen Li, Atsushi Nitanda, Taiji Suzuki, Denny Wu, Ji Xu

Keywords Paper

Keywords Paper

Keywords Paper

Valentin Thomas, Fabian Pedregosa, Bart van Merriënboer and
Pierre-Antoine Manzagol, Yoshua Bengio, Nicolas Le Roux

Keywords Paper