Search papers, labs, and topics across Lattice.
This paper establishes a verifiable regularity criterion for conditional expectation operators (CEOs) and conditional mean embeddings (CMEs), focusing on their mapping properties between function spaces. By characterizing these properties through the regularity of the Radon鈥揘ikodym density of the conditional law, the authors provide a sufficient condition for CEOs to be bounded and Hilbert鈥揝chmidt, particularly in reproducing kernel Hilbert spaces (RKHS) norm-equivalent to Sobolev spaces. The findings have practical implications for validating CME representations and error bounds in nonparametric regression, Bayesian inverse problems, and Koopman operator theory, demonstrating that classical regularity results can be leveraged across these domains.
A simple regularity condition can unlock powerful validation techniques for conditional mean embeddings across diverse mathematical applications.
Conditional expectation operators (CEOs) and their associated conditional mean embeddings (CMEs) play a central role across applied mathematics and machine learning, appearing in nonparametric regression, Bayesian inverse problems, and Koopman operator theory. A fundamental question is when a CEO maps a function space on $\mathcal{Y}$ into a prescribed function space on $\mathcal{X}$, particularly a reproducing kernel Hilbert space (RKHS). We show that such mapping properties are characterized by the regularity of the Radon--Nikodym density of the conditional law, and establish a simple, verifiable sufficient condition under which the CEO is bounded and Hilbert--Schmidt. For RKHSs norm-equivalent to Sobolev spaces, this condition reduces to Sobolev regularity of the conditional density. The result yields a direct route to validate CME representations and error bounds for Galerkin-type and CME-based estimators. We verify the regularity condition in three settings: nonparametric regression, Bayesian inverse problems, and Koopman operator theory for stochastic dynamical systems. We show in each case that classical regularity results on the underlying probabilistic model imply the required mapping properties. The resulting framework offers a unified perspective on conditional expectation operators across probability, operator theory, kernel methods, and stochastic dynamics.
