A Measure-Theoretic Approach to Kernel Conditional Mean Embeddings

Abstract

We present an operator-free, measure-theoretic approach to the conditional mean embedding (CME) as a random variable taking values in a reproducing kernel Hilbert space. While the kernel mean embedding of unconditional distributions has been defined rigorously, the existing operator-based approach of the conditional version depends on stringent assumptions that hinder its analysis. We overcome this limitation via a measure-theoretic treatment of CMEs. We derive a natural regression interpretation to obtain empirical estimates, and provide a thorough theoretical analysis thereof, including universal consistency. As natural by-products, we obtain the conditional analogues of the maximum mean discrepancy and Hilbert-Schmidt independence criterion, and demonstrate their behaviour via simulations.

References in corpus (14)

• Equivalence of distance-based and RKHS-based statistics in hypothesis testing
• A Kernel Test of Goodness of Fit
• Learning from Distributions via Support Measure Machines
• Operator-valued Kernels for Learning from Functional Response Data
• Modelling transition dynamics in MDPs with RKHS embeddings
• Towards a Learning Theory of Cause-Effect Inference
• Characteristic and Universal Tensor Product Kernels
• Conditional Generative Moment-Matching Networks
• A Rigorous Theory of Conditional Mean Embeddings
• Causal Inference via Kernel Deviance Measures
• Hilbert Space Embeddings of POMDPs
• Regularised Least-Squares Regression with Infinite-Dimensional Output Space
• On Distance and Kernel Measures of Conditional Independence
• Consistent Kernel Mean Estimation for Functions of Random Variables

Cited by in corpus (11)

• A Rigorous Theory of Conditional Mean Embeddings
• Kernel Partial Correlation Coefficient -- a Measure of Conditional Dependence
• Singular Value Decomposition of Operators on Reproducing Kernel Hilbert Spaces
• SOCKS: A Stochastic Optimal Control and Reachability Toolbox Using Kernel Methods
• When Invariant Representation Learning Meets Label Shift: Insufficiency and Theoretical Insights
• Conditional Distributional Treatment Effect with Kernel Conditional Mean Embeddings and U-Statistic Regression
• Kernel Autocovariance Operators of Stationary Processes: Estimation and Convergence
• Nonparametric approximation of conditional expectation operators
• Calibration tests beyond classification
• Exact Distribution-Free Hypothesis Tests for the Regression Function of Binary Classification via Conditional Kernel Mean Embeddings
• Sobolev Norm Learning Rates for Conditional Mean Embeddings